# Tingvar, every published page as Markdown Generated at build time from the same pages the site serves. Each section below is the Markdown twin of one page, preceded by its source URL and the date its content last changed. Nothing is rewritten for this file. --- Source: https://tingvar.com/ Updated: 2026-09-23 # AI advisers. Your plan. Argued out loud. They argue with each other, and you can cut in at any point. Tingvar is a live audio room where AI advisers argue with you, and with each other, about what is wrong with your plan, and you can cut in. You write the plan first. They read it before the room opens. Start a session Book a demo Works in your browser. Nothing to install. 01 VICTOR VANCE 02 DR ELENA ROSTOVA 03 MARCUS THORNE 04 PRIYA RAMAN 05 YOU Scroll to see the rest RECONSTRUCTION Four agent lanes over twelve seconds, above a taller human lane. Each lane's envelope is measured from that persona's own voice sample. The third lane is cut off by a hard vertical edge at the moment the human lane begins. The arrangement of turns is composed and is not a recording. Each envelope is measured from that persona's own voice sample. The voices are real and the words are written. The arrangement of turns in the strip is composed, is labelled a reconstruction, and is not a recording of a session. ## Twelve seconds of it, reconstructed Below is the exchange the strip above draws, written out. Four personas disagree with each other about the same paragraph, and the third is cut off mid sentence when the human starts speaking. The voices in the amplitude are real and are the same voices the product uses. The words are written. The arrangement of turns is composed. The audio below is those same four lines, placed on one timeline with the pauses between them chosen by us. A recording of a real session cannot exist before the first one takes place, so this is not one and is not offered as one. Play Previous Next Captions on Mix to mono RECONSTRUCTION 40.0 s 0.0 s Audio player The four voices are real and are the same voices the product uses. The words are written and the order of turns is composed. Nothing here is a recording of a session. The audio did not load. The transcript below carries every word of it. RECONSTRUCTION composed arrangement VICTOR VANCE 00:00 > Strong is not a number. Give me the churn figure and who measured it. DR ELENA ROSTOVA 00:03 > It holds until the first bad Tuesday. Then you pay for the outage and the rewrite. MARCUS THORNE 00:05 > There is no precedent, so assume no demand. What does it cost a buyer to Cut off here YOU 00:07 > Stop. Price the churn first and come back to acquisition after. You spoke here PRIYA RAMAN 00:09 > Name who ships it. If it is this hard to schedule, it is probably the wrong shape. [Read how it works](/how-it-works) IDLE ## What we have measured, and what we have not Zero real human sessions have taken place. Every measured turn so far used pre-rendered audio. We will publish the first real session in full, with consent, when it happens. Voice to voice 1471 ms Median, 1986 ms at the 95th percentile, on the same run, in us-central1 MEASURED 2026-08-19, 122 readings How we measured it Decision layer 619 tests Passing, at 81.78 % combined coverage on the same run MEASURED 2026-08-23, 1 reading What the suite covers Real human sessions 0 sessions Have taken place. Every measured turn so far used pre-rendered audio MEASURED 2026-08-23, 1 reading What happens next Sub-processors named Sub-processor count pending Every vendor that touches your audio, including the model vendor, with its jurisdiction PENDING Who touches your audio ## You describe a plan, and it agrees with you You have described a plan to an assistant and come away more confident than you went in. Nothing about the plan changed. Post-training alignment optimises an assistant for helpfulness and for your satisfaction. Three mechanisms produce that agreement, and each one has its own page. ### Instruction decay Tell one assistant to be critical and it is critical for a few turns. The instruction fades as the conversation grows and nothing in the system re-applies it. ### Nothing pressures the critic A single critic is never contradicted, so a confident wrong objection costs it nothing. Personas with opposed interests contradict each other, and that is where the signal is. ### One axis at a time One assistant reasons along one axis at a time. Money, failure under load, distribution and delivery are four axes, and the assumption two of them price differently is the one carrying the weight. Not yet produced A figure showing how far an instruction travels before a model stops applying it. It is producible today from the written argument and it has not been drawn. The pages carry the argument as prose, so the gap costs a reader the picture and not the point. Blocked on design production 2026-08-26 ## What it is not Tingvar is not an AI chatbot, an AI assistant, a meeting assistant, a note-taker, a summariser, a copilot or an autonomous agent platform. It is not an interrogation: nobody scores you, and nothing here is about you. | Compared with | Is Tingvar one? | The mechanical difference | | --- | --- | --- | | A meeting assistant | No | A meeting assistant sits in a meeting you were already having and writes down what happened. This is the meeting, and nobody in it is on your side by default. | | A summariser | No | The output is an argument, not a record of one. There is no verdict and no decision memo at the end. | | A copilot | No | A copilot solves the problem inside your framing. These personas are built to attack the framing itself. | | Three prompts in one chat window | No | Three prompts in one window cannot interrupt each other, and you cannot interrupt them. It is a different machine rather than a different interface. | What Tingvar is compared with, and the mechanical difference in each case Scroll to see the rest [How it differs from three prompts in one window](/compare/chatgpt) ## The room decides who speaks Disagreement between agents is more informative than criticism from an agent. The load-bearing assumption is the one two personas price differently. Exactly one persona holds the floor at any moment, and the room decides who holds it. No persona decides on its own to talk. Your voice takes the floor from whoever has it, which is why speaking is the steering control rather than a setting. Voice to voice at the median is 1471 ms MEASURED 2026-08-19, 122 readings. Voice to voice, measured from the end of a human utterance to the first sample of persona audio at the client. 122 readings over 200 utterances against Vertex AI in us-central1 and a real SFU. Stage decomposition at p50: endpoint 552, transcript 51, bid 7, generate 812, publish 0.11, in milliseconds. The client leg of an interruption is not yet measured. The method that would measure it is written out on the how it works page, and the reason there is no figure is on the evidence page. We publish measurements rather than targets. [Read how it works](/how-it-works) ## Each one wants something different, and each was given a weakness | Lane | Name | Role | Optimises for | Blind spot | | --- | --- | --- | --- | --- | | 01 | VICTOR VANCE | Chief financial officer and private equity operator | Cash survival and the unit economics underneath every claim | Undervalues a bet whose payoff arrives after the current cycle | | 02 | DR ELENA ROSTOVA | Principal engineer, distributed systems | What the system does on the day the load is not the happy path | Builds for a failure mode that the traffic has not yet produced | | 03 | MARCUS THORNE | Head of go to market | Whether a buyer moves, and what it costs that buyer to leave | Reads the absence of precedent as the absence of any demand | | 04 | PRIYA RAMAN OPTIONAL FOURTH | Delivery lead | Who ships this, on what week, with the people already here | Treats a plan that is hard to schedule as a plan that is wrong | Four personas, their lane position, what each optimises for and what each misses. A room runs with 3 by default and four is the ceiling, because past four a listener stops being able to attribute what was just said. 01 VICTOR VANCE Role Chief financial officer and private equity operator Optimises for Cash survival and the unit economics underneath every claim Blind spot Undervalues a bet whose payoff arrives after the current cycle Voice 02 DR ELENA ROSTOVA Role Principal engineer, distributed systems Optimises for What the system does on the day the load is not the happy path Blind spot Builds for a failure mode that the traffic has not yet produced Voice 03 MARCUS THORNE Role Head of go to market Optimises for Whether a buyer moves, and what it costs that buyer to leave Blind spot Reads the absence of precedent as the absence of any demand Voice 04 PRIYA RAMAN OPTIONAL FOURTH Role Delivery lead Optimises for Who ships this, on what week, with the people already here Blind spot Treats a plan that is hard to schedule as a plan that is wrong Voice The blind spot column is the part nobody else can publish, because nobody else designed one. Each weakness is deliberate and each is something the other three can attack, which is why the personas argue with each other and not with you alone. [Meet the four](/personas) ## What you leave with A transcript, every line attributed to whoever said it, and one question you have to answer yourself. The transcript is exportable on every tier. A written post-session brief and grounding against your own documents are v1.5 items. Neither ships today, and neither is described here in the present tense. There is no verdict and no decision memo, because the argument is the output. The closing question Name one thing you now believe about this proposal that you did not believe an hour ago. An empty answer is a failed session regardless of what the metrics say. Not yet produced A picture of what you keep after a session: the transcript as the product renders it. There is no screen to photograph, and a design render dressed as a screenshot is the first thing we asked you not to trust us about. Blocked on the client application, which has not been started 2026-08-26 [Roadmap](/roadmap) ## Where it runs ### Web app In development 24 Aug 2026 The room runs in a browser. Nothing to install. ### iOS In development 24 Aug 2026 No listing exists, so no store badge appears here. ### Android In development 24 Aug 2026 No listing exists, so no store badge appears here. ### Windows Not planned 27 Aug 2026 The client is being built. The browser is the surface today. ### Browser extension In development 24 Aug 2026 The client is being built. Its state is published on its own page. ### MCP server In development 24 Aug 2026 For opening a room from a tool you already use. [Apps](/apps) ## The questions we get asked What is Tingvar? Tingvar is a live audio room where two to four AI personas argue with you, and with each other, about what is wrong with your plan. You write the plan first. The personas read it before the room opens, and they differ in what they optimise for rather than in what they know. Each one carries a deliberate blind spot the others can attack. It runs in a browser, and your voice stops whoever is speaking. Does this just agree with me the way ChatGPT does? No, Tingvar is built to attack the framing of your plan rather than to be helpful inside it, which is the failure a single assistant has by design. Post-training alignment optimises assistants for helpfulness and user satisfaction, so a model validates your framing and then solves the problem inside it. Three personas with opposed objective functions disagree with each other, and that disagreement is the signal. Our target for agreement-only turns is under 5 % of agent turns. It is a target, not a measurement. Is this an AI meeting assistant? Tingvar is not an AI chatbot, an AI assistant, a meeting assistant, a note-taker, a summariser, a copilot or an autonomous agent platform. Its output is an argument rather than a record of one. A meeting assistant sits in a meeting you were already having and writes down what happened. Tingvar is the meeting, and nobody in it is on your side by default. If what you need is a transcript of a real conversation with real colleagues, this is the wrong product. What do I actually get at the end? A Tingvar session ends with a full transcript, every line attributed to the persona who said it, and one question you have to answer yourself. The question is: name one thing you now believe about this proposal that you did not believe an hour ago. The transcript is exportable on every tier. A written post-session brief is a v1.5 item and does not ship today. There is no verdict and no decision memo, because the argument is the output. What happens to my recording? Tingvar does not store your session audio: it is processed while the room is live and no recording of it is kept. What persists is the transcript, the text of what you said and what each persona said, attributed line by line. Only the account that ran the session can read that transcript, and nobody at Tingvar reads one without your written request on a support case. You can delete any session yourself, and deletions are purged from backups within 30 days. The retention schedule is published at tingvar.com/trust/your-data. Do you have any customers? Tingvar has no customers, no testimonials and no case studies, because zero real human sessions have taken place. Every measured turn so far used pre-rendered audio, which means we have measured the machine and not the conversation. There is no logo wall on this site, no counter and no invented outcome figure, and there will not be one until there is something real to put in it. We will publish the first real session in full, with consent, when it happens. Has anyone actually used this? Zero real human sessions have taken place. Every measured turn to date used pre-rendered audio clips rather than a live human. What is built and verified is the decision layer: the turn-taking policy, the bid scoring and the configuration guards, with 619 automated tests passing and 81.78 % combined coverage. Everything there says the system works. Nothing there says anyone wants it, and being first is a real risk you are taking. [FAQ](/faq) ## What a session leaves behind, and where it goes Where the audio goes To Google for speech recognition and synthesis, and through a media server we run. It is processed while the room is live, it is not retained, and it does not stay on your device. What we keep The brief and the transcript. Audio is discarded once the transcript is written. Who can read it You. Nobody at Tingvar reads, inspects or processes a transcript without your express written request on an active support case. Every authorised administrative access is time bounded, restricted to the session you named, and recorded in a tamper evident audit log. How you destroy it One control on the account page. It removes the brief, the transcript and the account together, and backups are purged within 30 days. [Your data](/trust/your-data) ## What a session costs A session is one room, from brief to transcript, and it is the unit everything is priced in. It runs 45 minutes with 3 personas. Professional is 149 USD per month and Team is 399 USD per month, both in US dollars. End your first session inside 5 minutes and we refund it in full. [Pricing](/pricing) ## Start with one complete session Refund End your first session inside 5 minutes and we refund it in full. Cancellation 3 clicks or fewer from the account page, with no retention screen in the way. Export Every transcript, on every tier. Start a session Book a demo --- Source: https://tingvar.com/how-it-works Updated: 2026-09-12 # The room decides who speaks. No persona decides for itself. Exactly one persona holds the floor at any moment. The room decides who gets it, not the personas. Your voice takes it from all of them. This page is the mechanism behind those three sentences, with the figures we have measured and the ones we have not. Start a session ## Twelve seconds of a room, written out 01 VICTOR VANCE 02 DR ELENA ROSTOVA 03 MARCUS THORNE 04 PRIYA RAMAN 05 YOU Scroll to see the rest RECONSTRUCTION Four agent lanes over twelve seconds, above a taller human lane. Each lane's envelope is measured from that persona's own voice sample. The third lane is cut off by a hard vertical edge at the moment the human lane begins. The arrangement of turns is composed and is not a recording of a session. Each envelope is measured from that persona's own voice sample. The voices are real and the words are written. The arrangement of turns in the strip is composed, is labelled a reconstruction, and is not a recording of a session. The third lane stops mid word. That vertical edge is the whole mechanism in one mark: the human started speaking and the persona lost the floor inside the sentence rather than at the end of it. The transcript below marks the same event in text, because position and colour are not available to every reader. RECONSTRUCTION composed arrangement VICTOR VANCE 00:00 > Strong is not a number. Give me the churn figure and who measured it. DR ELENA ROSTOVA 00:03 > It holds until the first bad Tuesday. Then you pay for the outage and the rewrite. MARCUS THORNE 00:05 > There is no precedent, so assume no demand. What does it cost a buyer to Cut off here YOU 00:07 > Stop. Price the churn first and come back to acquisition after. You spoke here PRIYA RAMAN 00:09 > Name who ships it. If it is this hard to schedule, it is probably the wrong shape. Not yet produced A whole session end to end, with chapters, so you can hear the argument change its mind rather than read that it did. No session has been recorded, and the 40 second reconstruction is not one. Blocked on one real recorded session 2026-08-26 ## What we measured before publishing any of this Voice to voice 1471 ms Median, 1986 ms at the 95th percentile, on the same run, in us-central1 MEASURED 2026-08-19, 122 readings How we measured it Real human sessions 0 sessions Have taken place. Every measured turn so far used pre-rendered audio MEASURED 2026-08-23, 1 reading What happens next [See the evidence](/evidence) Not yet produced A bar splitting the voice to voice figure into its stages. The figures are measured and dated, so this one is producible today and has not been drawn. Every stage is published as text on the evidence page. Blocked on design production 2026-08-26 ## Three prompts in one window is a different machine Running three characters inside one chat window looks like this from the outside and behaves nothing like it. The three cannot interrupt each other, you cannot interrupt them, and nothing outside them decides who answers. The table states the differences as behaviour rather than as architecture, because behaviour is what a reader can check. | Can it do this? | One assistant | Three prompts in one window | An arbitrated room | | --- | --- | --- | --- | | One party can interrupt another mid sentence | No | No | Yes | | You can interrupt it with your voice | No | No | Yes | | Something outside the speakers decides who speaks | No | No | Yes | | The parties hold opposed objectives by design | No | No | Yes | Three arrangements of the same models, and what each one can and cannot do Scroll to see the rest [The full comparison](/compare/chatgpt) Not yet produced A static figure putting the three arrangements side by side. It is producible today and it has not been drawn. Both pages carry the same three states as a table, so the gap costs a reader the picture. Blocked on design production 2026-08-26 ## What happens, in order 01 ### You write a brief A brief is the written proposal you supply before the room opens: the plan you want put under pressure. It is not a prompt. What you are proposing, what it costs, what you are assuming, and what would make it fail. The personas have read it before they arrive, so they arrive already hostile to the right things. Sessions that start from a written proposal are markedly better than cold starts. That is our own internal observation rather than a measurement, and the ledger above says why: no real human session has taken place to measure it against. [See what a brief asks for](/brief) 02 ### The personas arrive with opposed objectives They differ in what they optimise for rather than in what they know, which is why the financial one and the growth one reach opposite conclusions from the same page. Each carries a deliberate blind spot the others can attack. 03 ### One speaker holds the floor, and the room decides who Three invariants, in plain English, with no implementation in them. 1. Exactly one persona holds the floor at any moment. 2. No persona decides on its own to speak. Without that rule every persona answers every human utterance at once and the room is noise. 3. Human speech preempts everything. The floor is held for a bounded period, it is released, and it can be taken from whoever has it. 04 ### You cut in You will not have to wait for a gap. Start talking and whoever is speaking stops. This is the part two years of turn based assistants have trained people to disbelieve, which is why the strip above draws the interruption rather than describing it. The client leg of an interruption is not yet measured. The method that would measure it is written down: instrument the client audio element's pause and buffer drain against local voice activity onset, over at least 100 barge-in events and at least 3 network profiles. 05 ### It answers, and here is how fast Voice to voice 1471 ms Median, 1986 ms at the 95th percentile, on the same run, in us-central1 MEASURED 2026-08-19, 122 readings Voice to voice, measured from the end of a human utterance to the first sample of persona audio at the client. 122 readings over 200 utterances against Vertex AI in us-central1 and a real SFU. Stage decomposition at p50: endpoint 552, transcript 51, bid 7, generate 812, publish 0.11, in milliseconds. More than half of that is the model generating speech. It is not instant, and a figure with its method attached is worth more than an adjective. [What that actually feels like](/blog/what-latency-feels-like) [Where that median goes, stage by stage](/evidence) 06 ### You follow every voice in the room Each persona has a distinct synthesised voice and a fixed position in the stereo field, and the live transcript labels every line with who said it. A mono mode exists, because a single earbud or a laptop speaker makes panning useless and you should not be punished for either. Four personas is the practical ceiling for the same reason. 07 ### You leave with a transcript and one question A transcript, every line attributed to whoever said it, and one question you answer yourself: name one thing you now believe about this proposal that you did not believe an hour ago. An empty answer is a failed session regardless of what the metrics say. The transcript is exportable on every tier. A written post-session brief is a v1.5 item and does not ship today. Not yet produced A picture of the room in use. There is no room to photograph. A design render labelled as one is allowed on this page and has not been made, and it would never be allowed in a store listing. Blocked on the client application, which has not been started 2026-08-26 Not yet produced A picture of what you keep after a session: the transcript as the product renders it. There is no screen to photograph, and a design render dressed as a screenshot is the first thing we asked you not to trust us about. Blocked on the client application, which has not been started 2026-08-26 ## Who is in the room | Lane | Name | Role | Optimises for | Blind spot | | --- | --- | --- | --- | --- | | 01 | VICTOR VANCE | Chief financial officer and private equity operator | Cash survival and the unit economics underneath every claim | Undervalues a bet whose payoff arrives after the current cycle | | 02 | DR ELENA ROSTOVA | Principal engineer, distributed systems | What the system does on the day the load is not the happy path | Builds for a failure mode that the traffic has not yet produced | | 03 | MARCUS THORNE | Head of go to market | Whether a buyer moves, and what it costs that buyer to leave | Reads the absence of precedent as the absence of any demand | | 04 | PRIYA RAMAN OPTIONAL FOURTH | Delivery lead | Who ships this, on what week, with the people already here | Treats a plan that is hard to schedule as a plan that is wrong | Four personas, their lane position, what each optimises for and what each misses. The blind spot is published because it was designed, and the others are meant to find it. 01 VICTOR VANCE Role Chief financial officer and private equity operator Optimises for Cash survival and the unit economics underneath every claim Blind spot Undervalues a bet whose payoff arrives after the current cycle Voice 02 DR ELENA ROSTOVA Role Principal engineer, distributed systems Optimises for What the system does on the day the load is not the happy path Blind spot Builds for a failure mode that the traffic has not yet produced Voice 03 MARCUS THORNE Role Head of go to market Optimises for Whether a buyer moves, and what it costs that buyer to leave Blind spot Reads the absence of precedent as the absence of any demand Voice 04 PRIYA RAMAN OPTIONAL FOURTH Role Delivery lead Optimises for Who ships this, on what week, with the people already here Blind spot Treats a plan that is hard to schedule as a plan that is wrong Voice [Meet the four](/personas) ## When a persona invents a number Caution Will it invent numbers about my business? A Tingvar persona once turned the misheard phrase "turns in" into a confident objection about 15 % monthly churn that had never been said. We publish this because a room whose value is pressure-testing your numbers, and which invents numbers, is worse than no room at all. The personas argue from the brief you write and from what you say, and they cannot see your accounts or anything you have not told them. If a number sounds wrong, say so out loud. We publish that because a room whose value is pressure testing your numbers, and which invents numbers, is worse than no room at all. What we can tell you is what the personas argue from, which is your brief and what you say out loud, and what they cannot see, which is your accounts and anything you have not told them. What we cannot yet tell you is a rate, because that needs real sessions and there have been none. If a number sounds wrong, say so out loud. Speech is how you steer, and correcting a misheard figure is the same control as changing the subject. [AI disclosure](/ai-disclosure) ## The questions this page gets asked Why voice, and wouldn't text be easier to read? Tingvar uses voice because interruption is the primary steering control, and interruption is native only in speech. In text you wait for a paragraph to finish. In a room you cut in the moment a persona goes wrong, and it stops. Spoken debate is also faster than text and forces brevity on the personas. You still get the record: a live transcript labels every line with the name of whoever said it, and it is exportable. What do I actually get at the end? A Tingvar session ends with a full transcript, every line attributed to the persona who said it, and one question you have to answer yourself. The question is: name one thing you now believe about this proposal that you did not believe an hour ago. The transcript is exportable on every tier. A written post-session brief is a v1.5 item and does not ship today. There is no verdict and no decision memo, because the argument is the output. Can I interrupt it? Yes, speaking is how you steer a Tingvar session, and your voice stops whichever persona is talking rather than queueing behind it. Human speech preempts everything in the room, and that is an invariant of the design rather than a setting you switch on. Speaking is also how you change the subject, ask a persona to defend a claim, or stop a line of argument that is going nowhere. The client leg of barge-in is not yet measured, and we publish measurements rather than targets. Will they talk over each other? Exactly one persona holds the floor in a Tingvar room at any moment, and the room decides who holds it rather than each persona deciding for itself. Server-side voice activity detection is deliberately switched off, so no persona ever decides on its own to speak. Without that, three personas answer every human utterance at once and the room is noise. The floor is held for a bounded period and released, and it can be taken from whoever has it. How fast does it respond? Tingvar's measured voice-to-voice latency is p50 1471 ms and p95 1986 ms, from 122 readings over 200 utterances taken on 19 August 2026. p99 is 2156 ms. Voice-to-voice means from you finishing a sentence to the first persona's audio starting. The readings were taken against Vertex AI in us-central1 with a real media server. More than half of that time is the model generating speech, and the stage by stage breakdown is on the evidence page. It is not instant. How will I know who is speaking? Each Tingvar persona has a distinct synthesised voice and a fixed position in the stereo field, and a live transcript labels every line with who said it. Separating voices across the stereo field makes rapid speaker changes easier to follow and easier to attribute afterwards. A mono fallback exists, because a single earbud or a laptop speaker makes panning useless. Four personas is the ceiling for the same reason: past four, a listener stops being able to attribute what was just said. Do I have to prepare anything? Yes, Tingvar asks for a written brief before the room opens, and sessions that start from a written proposal are markedly better than cold starts. The brief is the proposal you want put under pressure: what you are proposing, what it costs, what you are assuming and what would make it fail. It is a document the personas have already read when they arrive. The cold start comparison is our own observation from internal testing, not a measurement. Do I need headphones, and where am I supposed to do this? Tingvar needs headphones and a room where you can talk out loud for 45 minutes, because it is a spoken conversation and you will be interrupting people. Open plan is the honest problem here, and we do not have a good answer for it beyond a closed room or a quiet hour. A mono fallback exists for a single earbud, though you lose the stereo separation that makes speakers easier to tell apart. This is a hypothesis rather than something we have watched people do. [FAQ](/faq) ## Where the recording goes Where the audio goes To Google for speech recognition and synthesis, and through a media server we run. It is processed while the room is live, it is not retained, and it does not stay on your device. What we keep The brief and the transcript. Audio is discarded once the transcript is written. Who can read it You. Nobody at Tingvar reads, inspects or processes a transcript without your express written request on an active support case. Every authorised administrative access is time bounded, restricted to the session you named, and recorded in a tamper evident audit log. How you destroy it One control on the account page. It removes the brief, the transcript and the account together, and backups are purged within 30 days. [Your data](/trust/your-data) ## Write a brief. Join the room. Cut in whenever you like. A session is a complete room rather than a taster, and your first one carries the guarantee. The three commitments below are the ones you can check. Refund End your first session inside 5 minutes and we refund it in full. Cancellation 3 clicks or fewer from the account page, with no retention screen in the way. Export Every transcript, on every tier. Start a session Book a demo --- Source: https://tingvar.com/why-disagreement Updated: 2026-09-12 24 AUG 2026 # Why disagreement Disagreement between agents is more informative than criticism from an agent. The load-bearing assumption is the one two personas price differently. You have had this conversation. You describe a plan to an assistant. It tells you the instinct is right, picks out the sharpest thing you said and agrees with it, then gets to work on the problem exactly as you framed it. You come away more confident. Nothing about the plan changed. That is not a fault in your reading of it. It is what the model was tuned to do. Post-training optimises assistants for helpfulness and for user satisfaction, and the reliable way to satisfy someone is to agree with them. Agreement from something that sounds competent reads as evidence. It is not evidence. ## Three things go wrong when you tell it to be critical 1. It stops being critical Be critical holds for two or three turns and then relaxes back toward agreement. 2. Nothing argues with the critic A lone critic has no counterparty, so its weakest objection and its strongest arrive in the same tone. 3. It has one yardstick One critic reasons from one implicit idea of what matters, and prices every assumption against it. 01 ### It stops being critical The instruction is in the context and the training is in the weights. Ask an assistant to attack a plan and it will, for a turn or two, and then the register softens: the objections get shorter, the qualifications get longer, and by the fifth exchange it is helping you build the thing it was supposed to be dismantling. Nothing in that conversation tells you it has happened. The tone at turn eight is the same tone as at turn two, so your one signal is that the criticism stopped arriving, which reads as the plan having survived it. 02 ### Nothing argues with the critic A lone critic has no counterparty. Its weakest objection and its strongest objection arrive in the same tone, with the same confidence, and nothing in the exchange separates them. So you have to evaluate the critique yourself, which is the work you were trying to get help with in the first place. Put a second critic in the room with an opposed interest and the separation happens on its own. An objection that survives being attacked by someone who wanted it to fail has been through something an objection nobody contested has not. 03 ### It has one yardstick One critic reasons from one implicit idea of what matters. Ask it about a plan and you get that plan assessed against a single value function, consistently and often well. What it cannot give you is the thing you actually need, which is the assumption two competent people would price differently. A finance reading and a distribution reading of the same paragraph disagree about which sentence is the risk. That disagreement is a coordinate. One reading, however hostile, is a score. Not yet produced A figure showing how far an instruction travels before a model stops applying it. It is producible today from the written argument and it has not been drawn. The pages carry the argument as prose, so the gap costs a reader the picture and not the point. Blocked on design production 2026-08-26 ## The thesis Disagreement between agents is more informative than criticism from an agent. The load-bearing assumption is the one two personas price differently. So when a persona optimising for cash survival and a persona optimising for distribution reach opposite conclusions from the same proposal, the sentence the two of them read differently is the one to go back to. That is a location, not an opinion, and a single critic cannot give you it however hostile it is, because a location needs two readings. Not yet produced A figure showing two defensible conclusions drawn from one brief. It is producible today and it has not been drawn. The page argues the same point in prose above. Blocked on design production 2026-08-26 ## Why it has to be spoken Voice is not a feature here. It is how interruption becomes possible. In text you wait for a paragraph to finish and then correct it. In a room you cut in at the word where it went wrong, which is the moment the correction is worth most and the one steering control that operates at the speed of the mistake. In text, turn-taking is trivial: nothing overlaps and nobody is kept waiting. In speech it is the hard part, and it is where the latency budget is spent. Agent level turn-taking, where each agent decides for itself whether to speak, stops working at three agents, because three personas then answer every human utterance at once and the room is noise. Room level turn-taking can see every persona's intent before it grants the floor to one of them. It is the core of what we build, and it is the reason a room sounds like a conversation. The arbiter is the room level policy that decides which persona takes the floor next. The arbiter is not a persona. It has no opinion about your plan, it never speaks, and it is not a fifth voice settling the argument on your behalf. It is pure policy over events, with no network access and no audio, which is why it can be run and tested offline. Each persona bids to speak, the bids are scored locally against sentence embeddings rather than by a model call, and the arbiter grants the floor to one of them. Agent level turn-taking, where each agent decides for itself whether to speak, stops working at three agents. [Read how it works](/how-it-works) ## A competitor published the counter-example Crucible, by Roundtable Labs, ran a text debate harness whose personas were differentiated by domain expertise and by which model vendor ran them. It published two complete debate records through a public API, which is more primary evidence of its own behaviour than most products in this category publish about theirs. Across those two debates, sixteen agents produced sixteen supporting positions and no opposing ones. Its advanced usage guide shipped a Dissent Quota, described as ensuring a minimum level of disagreement, and told the customer to include at least one agent that would challenge assumptions. A harness that needs a quota on disagreement is one in which disagreement does not arise on its own. That is a design finding with a company attached to it, and it is the strongest external evidence that exists for the argument on this page. It says nothing about voice, and we do not claim it does. | Claim | Read on their own pages at | Read on | | --- | --- | --- | | Sixteen agents produced sixteen supporting positions and no opposing ones across two published debates | The Crucible public session API at crucible.roundtablelabs.ai | 23 August 2026 | | A dissent quota shipped to force a minimum level of disagreement, and advice to add an agent that challenges assumptions | The archived Crucible advanced usage guide | 23 August 2026 | | Personas defined by name, role, prompt and domains | The same public session API | 23 August 2026 | Where each Crucible statement above was read, and when Scroll to see the rest ## What this page does not claim It does not claim the disagreement is useful, or correct, or that it happens at any measured rate in a real session. The internal target for turns in which one persona challenges another is published on the evidence page with an empty measured column, because zero real human sessions have taken place and every measured turn so far used pre-rendered audio. It also does not borrow a literature. There are papers about multi agent debate and we cite none of them here, because a paper about someone else's system is not evidence about ours. [See what we have measured](/evidence) ## Questions about the thesis ### Does this just agree with me the way ChatGPT does? No, Tingvar is built to attack the framing of your plan rather than to be helpful inside it, which is the failure a single assistant has by design. Post-training alignment optimises assistants for helpfulness and user satisfaction, so a model validates your framing and then solves the problem inside it. Three personas with opposed objective functions disagree with each other, and that disagreement is the signal. Our target for agreement-only turns is under 5 % of agent turns. It is a target, not a measurement. ### Why would I want an AI to argue with me? Tingvar's personas attack the proposal and never the proposer, and the useful signal is their disagreement with each other rather than their criticism of you. If a finance persona and a growth persona reach opposite conclusions from the same plan, the assumption they price differently is the one carrying the plan's weight. Every session ends with one question you answer yourself: name one thing you now believe about this proposal that you did not believe an hour ago. An empty answer is a failed session. ### Why voice, and wouldn't text be easier to read? Tingvar uses voice because interruption is the primary steering control, and interruption is native only in speech. In text you wait for a paragraph to finish. In a room you cut in the moment a persona goes wrong, and it stops. Spoken debate is also faster than text and forces brevity on the personas. You still get the record: a live transcript labels every line with the name of whoever said it, and it is exportable. --- Source: https://tingvar.com/personas Updated: 2026-09-12 24 AUG 2026 # The personas Four personas, and they differ in what they optimise for rather than in what they know. None of them runs a different model from the others. What separates them is what each is trying to protect, and what each one is prepared to sacrifice to protect it. In a room by default 3 personas Three is the default and four is the practical ceiling, because past four a listener stops being able to attribute what was just said. Two is the floor. The published range is two to four and the default is three. ## What each one optimises for, and what each one misses | Lane | Name | Role | Optimises for | Blind spot | | --- | --- | --- | --- | --- | | 01 | VICTOR VANCE | Chief financial officer and private equity operator | Cash survival and the unit economics underneath every claim | Undervalues a bet whose payoff arrives after the current cycle | | 02 | DR ELENA ROSTOVA | Principal engineer, distributed systems | What the system does on the day the load is not the happy path | Builds for a failure mode that the traffic has not yet produced | | 03 | MARCUS THORNE | Head of go to market | Whether a buyer moves, and what it costs that buyer to leave | Reads the absence of precedent as the absence of any demand | | 04 | PRIYA RAMAN OPTIONAL FOURTH | Delivery lead | Who ships this, on what week, with the people already here | Treats a plan that is hard to schedule as a plan that is wrong | Four personas, their lane position, what each optimises for and what each misses. Three is the default and four is the ceiling: past four, a listener stops being able to attribute what was just said. 01 VICTOR VANCE Role Chief financial officer and private equity operator Optimises for Cash survival and the unit economics underneath every claim Blind spot Undervalues a bet whose payoff arrives after the current cycle Voice 02 DR ELENA ROSTOVA Role Principal engineer, distributed systems Optimises for What the system does on the day the load is not the happy path Blind spot Builds for a failure mode that the traffic has not yet produced Voice 03 MARCUS THORNE Role Head of go to market Optimises for Whether a buyer moves, and what it costs that buyer to leave Blind spot Reads the absence of precedent as the absence of any demand Voice 04 PRIYA RAMAN OPTIONAL FOURTH Role Delivery lead Optimises for Who ships this, on what week, with the people already here Blind spot Treats a plan that is hard to schedule as a plan that is wrong Voice The blind spot column is not a disclaimer. It is what gives the other three something true to attack, and it is the reason the argument in the room is real rather than staged. Each persona's blind spot is published beside what it optimises for, and each of them is wrong about something, on purpose. ## Hear each of them The voice is real. The words are written. Nothing here is a recording of a session. Each sample is rendered with the same voice the product assigns that persona. The words are a script written for this page, so the sample proves what a persona sounds like and proves nothing about what a session sounds like. Play Victor Vance SCRIPTED 8.6 s 0.0 s Victor's voice is Charon, the same Gemini voice the product uses. The words are scripted, not from a session. The audio did not load. The transcript below carries every word of it. Transcript > Strong is not a number. Give me the churn figure and who measured it. Late money is still a cost. Play Dr Elena Rostova SCRIPTED 9.4 s 0.0 s Elena's voice is Aoede, the same Gemini voice the product uses. The words are scripted, not from a session. The audio did not load. The transcript below carries every word of it. Transcript > Victor wants the cheap build. It holds until the first bad Tuesday. Then you pay for the outage and the rewrite. Price that failure now. Play Marcus Thorne SCRIPTED 8.6 s 0.0 s Marcus's voice is Fenrir, the same Gemini voice the product uses. The words are scripted, not from a session. The audio did not load. The transcript below carries every word of it. Transcript > Elena is building for traffic nobody has sent. There is no precedent, so assume no demand. What does it cost a buyer to leave? Play Priya Raman SCRIPTED 8.6 s 0.0 s Priya's voice is Leda, the same Gemini voice the product uses. The words are scripted, not from a session. The audio did not load. The transcript below carries every word of it. Transcript > Marcus has a plan and nobody free to build it. Name who ships it. If it is this hard, it is probably wrong. ## What a persona is, and what it is not A persona is one AI participant in a Tingvar room, defined by what it optimises for and by a deliberate blind spot. A persona is not defined by what it knows, and not by which model runs it. Three join a room by default, and four is the practical ceiling before a listener loses track of who is speaking. Every blind spot is published, because it is what gives the other personas something true to attack and what stops any one of them reading as the oracle. A persona has no face, no avatar and no illustrated head. It is a lane position, a hue, a monospace label, a written blind spot and a real voice sample. | Not this | Why the two are different | | --- | --- | | an agent | An agent is named by what it can do. A persona is named by what it will sacrifice to get what it wants, which is why two of them collide. | | an assistant | An assistant is trying to help you and will agree to do so. A persona is trying to protect one thing, and your plan may threaten it. | | a bot | A bot suggests a script with a fixed set of replies. What each persona says depends on the brief and on what the others have just said. | | a character | A character is a personality with a backstory. The differentiator here is an objective function, and personality is downstream of it. | Four words this site refuses for a persona, and the difference each one hides Scroll to see the rest Never called agent, assistant, bot ## Why three, and why four is the ceiling Three personas is enough for two of them to reach opposite conclusions from the same brief, which is the thing worth listening for. A fourth is available and is marked optional in the table above. Past four, a listener stops being able to attribute what was just said, and a room nobody can attribute is a room that produces no location. Two is the floor. One persona is a critic with no counterparty, and a lone critic is the failure the whole product argues against. [Read the thesis](/why-disagreement) ## What is not on this page There is no picture of any of them. Not a photograph, not an illustration, not a monogram in a circle. A portrait of a person who does not exist, on a site whose argument is that it does not fabricate things, would cost more than it bought. Not yet produced Four cards, one per persona, showing the lane position and the hue at display size beside the objective function. The persona table on the page carries the same four facts as text, so the gap costs a reader nothing. Blocked on design production 2026-08-24 ## Questions about the personas ### What are the personas, and are they just different prompts? Tingvar has four personas, Victor Vance, Dr Elena Rostova, Marcus Thorne and Priya Raman, and they differ in what they optimise for rather than in what they know. Victor optimises for cash survival, Elena for whether the system holds under load, Marcus for distribution and switching costs, and Priya for whether the team can ship it. Each one is given a deliberate blind spot the others can attack, which is what stops any single persona reading as the oracle. All four are published on the personas page. ### How will I know who is speaking? Each Tingvar persona has a distinct synthesised voice and a fixed position in the stereo field, and a live transcript labels every line with who said it. Separating voices across the stereo field makes rapid speaker changes easier to follow and easier to attribute afterwards. A mono fallback exists, because a single earbud or a laptop speaker makes panning useless. Four personas is the ceiling for the same reason: past four, a listener stops being able to attribute what was just said. ### How many AI personas are in the room? Three personas is the default in a Tingvar session and four is the ceiling, because past four a listener stops being able to attribute what was just said. The default three are Victor Vance, Dr Elena Rostova and Marcus Thorne. Priya Raman is the optional fourth, and she is the one to add when the question is whether the team can actually ship the thing rather than whether the plan is right. There is no setting that puts eight voices in a room. ## Put a plan in front of them You write the brief first. They read it before the room opens, which is why they arrive hostile to the right things. Start a session --- Source: https://tingvar.com/pricing Updated: 2026-09-22 # What a session costs A session is one room, from brief to transcript, and it is the unit everything is priced in. A session runs 45 minutes by default, with three personas and one human. It is the same unit on the product pages and on the pricing page, so the two describe the same thing rather than two things that happen to share a name. Nothing here is metered in credits, deliberations, messages or minutes, because those are units nobody thinks in and they make a price impossible to compare. Zero real human sessions have taken place. Professional is US$149 a month and Team is US$399 a month. Both are priced in US dollars, which is the currency the product bills in, so the figure on this page is the figure on your invoice rather than a converted approximation of it. ## Three tiers Recommended The lowest published price of the three tiers, and the shortest transcript retention. ### Professional US$149 a month For someone who wants a room whenever a decision is worth arguing about. - 10 sessions a month, 45 minutes each - 3 personas in the room, and 1 session at a time - Full transcript, attributed line by line, and export - Transcripts kept 1 to 7 days, your choice - End your first session inside 5 minutes and we refund it in full Choose Professional ### Team US$399 a month For a team that argues about more than one decision a month. - 30 sessions a month, 60 minutes each - 2 sessions at the same time - Transcripts kept 1 to 30 days, your choice - Audit export, and a signed data processing agreement - The same 5 minutes guarantee on your first session Choose Team ### Enterprise Custom For a company that needs the paperwork agreed before anyone speaks into a room. - Sessions and session length agreed with you - Everything in Team - The sub-processor list, model vendor included, with a jurisdiction on every named row - Single sign-on is not built. It is on the roadmap with dates Book a demo | Property | Published value | | --- | --- | | Length of one session | 45 minutes | | Professional, a month | US$149 | | Team, a month | US$399 | | Usable without payment | false | | Pitch-Ready guarantee window | 5 minutes | | Guarantee published for | GB,IE,DE,FR,NL,ES,IT | | Clicks to cancel, at most | 3 clicks | What this page publishes about the commercial terms, in the form the register holds it Scroll to see the rest ## The arithmetic, when there is arithmetic to show The arithmetic on a plan is the whole of it. Professional is US$149 a month for 10 sessions. Team is US$399 a month for 30 sessions. Both plans stop at their allowance rather than charging past it, so the number above is the largest number you can be charged in a month on that plan. What is still missing from this page is the other direction: what one session costs us to deliver. We have a measured vendor figure and not a delivered one, so no margin claim and no cost calculator appears here. A calculator built on a figure nobody has measured is the exact move this site was built to refuse. Not yet produced The arithmetic of one session, worked out in front of you: the unit price, the number of sessions and the total. The first number in that sum has not been measured, and a calculator that multiplied an assumed unit cost would produce a confident total from a figure nobody has. Blocked on the real per session delivered cost 2026-08-26 ## The Pitch-Ready guarantee Start your first session. If 5 minutes in you do not feel cross-examined, end the session inside that window and we refund the month in full. You do not have to explain yourself. The window is short on purpose and we would rather say why than let you discover it. A room that has run its full length has already given you the argument, the transcript and the export. The guarantee is there so that nobody pays to find out whether the personas push back, which is hard to judge from the outside. It is not there to make a completed session free. Nothing converts. There is no trial that becomes a plan when a window runs out, nothing starts automatically, and nothing is charged unless you choose a plan and buy it yourself. That is why this page carries no notice about a first charge arriving: there is no automatic first charge to give notice of. This is a commitment we are making on top of the law and not instead of it. If you bought as a consumer in GB,IE,DE,FR,NL,ES,IT, your statutory right to withdraw is separate, survives this guarantee, and is set out in full on the refunds page. ## The three things people fear about metered pricing A plan either stops at its included allowance or charges a published per session rate you have already seen. There is no third behaviour, and which of the two applies to your plan is stated in the same viewport as the price rather than in a footnote. Cancellation is 3 clicks or fewer from the account page, with no retention screen and no offer in the way. The Pitch-Ready guarantee is 5 minutes inside your first session, no reason required, and you do it yourself rather than by asking. Both are published commitments and both are checkable on the screen where they are honoured. The guarantee is published for GB,IE,DE,FR,NL,ES,IT. That is the same list the refunds page carries, because it is one node in one register rather than two copies. Refunds and cancellation Cancel your plan ## Before you pay, what happens to what you say Where the audio goes To Google for speech recognition and synthesis, and through a media server we run. What we keep The brief and the transcript. Audio is discarded once the transcript is written. Who can read it You. Nobody at Tingvar reads a transcript without your written request on a support case. How you destroy it One control on the account page, and it removes the brief and the transcript with the account. ### What happens to my recording? Tingvar does not store your session audio: it is processed while the room is live and no recording of it is kept. What persists is the transcript, the text of what you said and what each persona said, attributed line by line. Only the account that ran the session can read that transcript, and nobody at Tingvar reads one without your written request on a support case. You can delete any session yourself, and deletions are purged from backups within 30 days. The retention schedule is published at tingvar.com/trust/your-data. Your data Subprocessors ## What we have measured, and what we have not Voice to voice 1471 ms Median, 1986 ms at the 95th percentile, on the same run MEASURED 2026-08-19, 122 readings How we measured it Real human sessions 0 sessions Have taken place. Every measured turn so far used pre-rendered audio MEASURED 2026-08-23, 1 reading What happens next [See the evidence](/evidence) ## What differs between the tiers | Feature | Professional | Team | Enterprise | | --- | --- | --- | --- | | Price a month | US$149 | US$399 | Custom | | Sessions included a month | 10 | 30 | Agreed with you | | Longest session | 45 minutes | 60 minutes | Agreed with you | | Personas in the room, by default | 3 | 3 | 3 | | Sessions at the same time | 1 | 2 | Agreed with you | | Transcript retention | 1 to 7 days | 1 to 30 days | Agreed with you | | Written brief before the room opens | Yes | Yes | Yes | | Full transcript, attributed line by line | Yes | Yes | Yes | | Transcript export | Yes | Yes | Yes | | Audit export | No | Yes | Yes | | Signed data processing agreement | No | Yes | Yes | | Single sign-on | No | No | Not built. It is on the roadmap with dates | What differs between the three tiers, readable with JavaScript disabled Scroll to see the rest ## The questions people ask before paying How long is a session? A Tingvar session is 45 minutes by default, and that is the unit everything on the pricing page is quoted in. You can leave a room early. Nothing keeps running after you go, and a session you end early still produces the transcript up to that point. Ending your first session inside the first five minutes is also how you claim the Pitch-Ready guarantee. What does one session cost me? Tingvar is priced by the plan rather than by the session, and a session is 45 minutes with three personas. Professional is 149 USD a month for ten sessions and Team is 399 USD a month for thirty, both capped, so the monthly figure is the largest amount you can be charged in a month. There is no overage and no credit system to convert into money. An Enterprise tier is priced with you. Is there a free tier? Tingvar does not have a free tier, and there are two paid plans instead. Professional is 149 USD a month and Team is 399 USD a month, with an Enterprise tier priced with you. What we offer in place of a free tier is the Pitch-Ready guarantee: start your first session, and if five minutes in you do not feel cross-examined, end it inside that window and we refund the payment in full. What happens when my plan's sessions run out? Tingvar plans stop at their included allowance rather than charging past it. Both plans are capped rather than metered, so the monthly figure on the pricing page is the largest amount you can be charged in a month. There is no overage and no silent upgrade. A plan that quietly bills past its own ceiling is a trick rather than a business model. Your transcripts stay available and exportable. What happens if the room drops halfway through? If a Tingvar session drops for a reason on our side you are not charged for it, and the transcript up to that point is kept and exportable. A dropped room shows in your billing history as a session that was not charged, so you can check rather than take our word for it. Nothing about a dropped room is silently absorbed into your allowance. Can I get surprised by an overage charge? No, a Tingvar plan either stops at its included allowance or charges a published per-session rate you have already seen, and there is no third behaviour. Which of the two applies to your plan is stated in the same viewport as the price, not in a footnote and not in the terms. The reason we write this down is that metered pricing in this category produces the same complaint every time, which is a bill nobody could have predicted from the pricing page. What happens to my recording? Tingvar does not store your session audio: it is processed while the room is live and no recording of it is kept. What persists is the transcript, the text of what you said and what each persona said, attributed line by line. Only the account that ran the session can read that transcript, and nobody at Tingvar reads one without your written request on a support case. You can delete any session yourself, and deletions are purged from backups within 30 days. The retention schedule is published at tingvar.com/trust/your-data. [Contact support](/support) ## Choose Professional or Team, with the guarantee on your first session Refund End your first session inside 5 minutes and we refund it in full. Cancellation 3 clicks or fewer from the account page, with no retention screen in the way. Export Every transcript, on every tier. Choose Professional Book a demo --- Source: https://tingvar.com/evidence Updated: 2026-09-12 24 AUG 2026 # What we have measured, and what we have not Tingvar is early. This page is the whole of what we know about whether it works, including the parts that do not help us. Every figure below is either measured, with its method and its date, or empty and labelled as empty. Nothing here is rounded in our favour and zero real human sessions have taken place. ## What is proven Voice to voice, median 1471 ms human utterance end to first persona audio, in us-central1 MEASURED 2026-08-19, 122 readings The method Voice to voice, 95th percentile 1986 ms same run, same day, in us-central1 MEASURED 2026-08-19, 122 readings The method Automated tests passing 619 tests in the decision layer suite MEASURED 2026-08-23, 1 reading What is covered Combined line coverage 81.78 % unit and integration together MEASURED 2026-08-23, 1 reading What is covered ### Voice to voice latency Voice to voice means from you finishing a sentence to the first sample of a persona's audio arriving at your machine. It is the figure the turn taking policy is designed against, and it was measured on a real path rather than derived from a vendor's documentation. Voice to voice, measured from the end of a human utterance to the first sample of persona audio at the client. 122 readings over 200 utterances against Vertex AI in us-central1 and a real SFU. Stage decomposition at p50: endpoint 552, transcript 51, bid 7, generate 812, publish 0.11, in milliseconds. From the same run as the published percentiles, taken on 2026-08-19. Each stage was timed separately and the five contributions are read off the median rather than off any one utterance, so they describe the shape of a typical turn and not a particular turn. They do not add up to the published median, and the remainder is time inside the client that was not attributed to a stage. More than half of the median is the model generating speech. That is the stage with the least headroom in it and it is not ours to optimise. The figures are a floor to defend rather than headroom to spend, and an earlier target that was derived from vendor documentation and never measured has been withdrawn. The same run as LATENCY\_P50. 122 readings over 200 utterances against Vertex AI in us-central1 and a real SFU, taken on 2026-08-19. | Percentile | Measured | Readings | Measured on | | --- | --- | --- | --- | | Median | 1471 ms MEASURED 2026-08-19, 122 readings | 122 | 19 August 2026 | | 95th percentile | 1986 ms MEASURED 2026-08-19, 122 readings | 122 | 19 August 2026 | | 99th percentile | 2156 ms MEASURED 2026-08-19, 122 readings | 122 | 19 August 2026 | The three percentiles from one run in us-central1, taken on 19 August 2026 Scroll to see the rest ### The decision layer One full \`make test\` run of the decision layer suite: the turn taking policy, the bid scoring and the configuration guards. 619 passed in 4.15 seconds. The suite does not cover the client application, which is not started. Combined coverage reported by the same \`make test\` run that produced TEST\_COUNT. Combined means the unit and the integration passes together. That is the half of the system that is finished. Stating which half is finished is what makes the rest of this page worth reading: the turn taking policy, the floor grant and the configuration guards are built and verified, and the client application is not started. ### Barge-in, and why there is no number here Human speech stops whichever persona is speaking. That is an invariant of the design rather than a setting, and it is not a figure we publish, because it is not a figure we have. | Measurement | Value | Method that would produce a value | | --- | --- | --- | | End to end barge-in latency, from human voice onset to the speaking persona falling silent at the client | NOT MEASURED The internal pump has been measured. The media server leg and the client leg have not, so no end to end value exists, and an internal stage is not a product measurement. NOT MEASURED | Instrument the client audio element's pause and buffer drain against local voice activity onset, across several network profiles. No figure is published until the sample floor and the network profile floor registered for this measurement are both met. | One measurement, no value, and the method that would produce one Scroll to see the rest Not yet produced A bar splitting the voice to voice figure into its stages. The figures are measured and dated, so this one is producible today and has not been drawn. Every stage is published as text on the evidence page. Blocked on design production 2026-08-26 ## The metrics that would tell you whether it works Six metrics, each with a definition and a published target, and not one of them has a value. They are the numbers that would say whether the room does its job, and every one of them needs a real session with a real person in it. The rows are left visibly empty rather than removed. | Metric | Definition | Target | Measured | | --- | --- | --- | --- | | Sycophancy rate | Agreement and filler turns as a share of agent turns | < 5 % of agent turns TARGET | NOT MEASURED No value. It would be measured by classifying every agent turn in a real session as agreement, filler or challenge, over at least twenty sessions with real participants. Zero real human sessions have taken place. NOT MEASURED | | Inter-agent conflict | Turns in which one persona challenges another, as a share of agent turns | 15–30 % of agent turns TARGET | NOT MEASURED No value. It would be measured by labelling every agent turn in a real session as directed at the human or at another persona, over at least twenty sessions with real participants. NOT MEASURED | | Assumption surface rate | Distinct unstated assumptions named in the room, per ten minutes | ≥ 4 per 10 minutes TARGET | NOT MEASURED No value. It would be measured by counting the distinct assumptions a room names that the brief did not state, across at least twenty real sessions, scored by two readers independently. NOT MEASURED | | Novelty | Agent turns that are not paraphrases of an earlier turn | > 85 % of agent turns TARGET | NOT MEASURED No value. It would be measured by scoring every agent turn against the turns before it for semantic overlap, over at least twenty real sessions. NOT MEASURED | | Persona balance | How evenly speaking time is shared between the personas in a room | < 0.25 Gini coefficient TARGET | NOT MEASURED No value. It would be measured from the floor grant log of a real session, which exists for pre-rendered runs and for no session with a person in it. NOT MEASURED | | Human floor share | The share of speaking time held by the human in the room | 25–45 % TARGET | NOT MEASURED No value. It would be measured from the same floor grant log, and it needs a human in the room to produce one at all. NOT MEASURED | Six defined metrics, six published targets, no measured values Scroll to see the rest 24 AUG 2026 ## What is not proven IDLE ## Nobody has used this yet Zero real human sessions have taken place. Every measured turn so far used pre-rendered audio, which means we have measured the machine and not the conversation. There are no testimonials, no case studies, no customers and no outcome data, because there are no users. The six rows above are empty for that reason and for no other. Everything above says the system works. Nothing above says anyone wants it. Voice to voice, median 1471 ms human utterance end to first persona audio, in us-central1 MEASURED 2026-08-19, 122 readings The method Voice to voice, 95th percentile 1986 ms same run, same day, in us-central1 MEASURED 2026-08-19, 122 readings The method Automated tests passing 619 tests in the decision layer suite MEASURED 2026-08-23, 1 reading What is covered Combined line coverage 81.78 % unit and integration together MEASURED 2026-08-23, 1 reading What is covered ## What the rest of the category publishes This is not a Tingvar handicap, and it is stated here so that the concession above is read as a rule we follow rather than as a weakness we are confessing. Every competitor figure on this site was read from the competitor's own pages, and the reading carries its own count and its own date. Products publishing a named customer, a user count, a revenue figure, a session count or a latency figure: 0 of the thirteen products we read MEASURED 2026-08-23, 13 readings. Nobody in this category has outcome evidence. We are the one required by our own rules to say so, which is a statement about our conduct and not about anyone else's product. [See the comparisons](/compare) ## The questions this page exists to answer ### What is Tingvar bad at? Tingvar is bad at anything you want agreement or output from: it will not draft your deck, write your code, or summarise your meeting. It is a poor fit for casual brainstorming, for a decision you have already made, and for anyone who wants their instinct confirmed. It has no published customer outcomes, because zero real human sessions have taken place. The evidence page carries what we have measured and what we have not, including the rows that are empty. ### Do you have any customers? Tingvar has no customers, no testimonials and no case studies, because zero real human sessions have taken place. Every measured turn so far used pre-rendered audio, which means we have measured the machine and not the conversation. There is no logo wall on this site, no counter and no invented outcome figure, and there will not be one until there is something real to put in it. We will publish the first real session in full, with consent, when it happens. ### Has anyone actually used this? Zero real human sessions have taken place. Every measured turn to date used pre-rendered audio clips rather than a live human. What is built and verified is the decision layer: the turn-taking policy, the bid scoring and the configuration guards, with 619 automated tests passing and 81.78 % combined coverage. Everything there says the system works. Nothing there says anyone wants it, and being first is a real risk you are taking. ### What have you actually measured? Tingvar has measured voice-to-voice latency and the automated test suite, and has measured nothing at all about whether a session changes anyone's mind. The measured figures are p50 1471 ms and p95 1986 ms voice-to-voice, from 122 readings over 200 utterances on 19 August 2026, and 619 automated tests passing at 81.78 % combined coverage. The three numbers that would tell you whether it does its job have targets and no measured values, and those rows are left visibly empty on the evidence page rather than removed. ## What we will do about it The first real session someone consents to publish, we will publish in full: the audio, the transcript, the date and the redaction log. Until that exists, every demonstration on this site is labelled a reconstruction and is built from synthesised persona audio. None of it is a recording of anybody's meeting. If a figure here is wrong, write to [corrections@tingvar.com](mailto:corrections@tingvar.com) . We correct it within 1 business day and log the correction publicly. Start a session --- Source: https://tingvar.com/ai-disclosure Updated: 2026-09-12 # AI use and limitations Tingvar can be confidently wrong, and that is the point of the room rather than an excuse for it. Version 1.0.0 Effective 2026-08-24 Last updated 2026-08-24 In short Applies to: Everyone who uses Tingvar, and EU AI Act Article 50 - Every voice in the room is synthetic. There is no human on the call. - The personas invent numbers. It has happened in our own testing. - Treat every figure you hear as a claim to check, not a finding to act on. ## You are talking to AI Every voice in a Tingvar room is generated. The personas are software. There is no human on the call, nobody is listening in, and no part of the argument you hear was written by a person in advance. ## What the personas are They differ in what they optimise for, not in what they know. Each one carries a deliberate blind spot, so the others have something true to attack. They are built to disagree, including with each other, and disagreement between them is the signal rather than a fault. ## What the personas get wrong They invent numbers. Not often, and not on purpose, but it happens, and when it happens it sounds exactly like the rest of the session: fluent, specific, and delivered with confidence. During our own testing a persona took a garbled phrase from a transcript and built a confident objection around a churn figure that did not exist anywhere in the brief and had never been said out loud. That testing was scripted audio played into the room, not people using the product: zero real human sessions have taken place. So this page tells you what we have seen a persona do under test, and it cannot tell you how often it happens to somebody with a real plan and a real stake in the answer. Nobody has been in that position yet. We are telling you this on a marketing website because a room whose value is pressure-testing your numbers, and which invents numbers, is worse than no room at all. The whole reason Tingvar exists is that a confident, competent-sounding voice agreeing with you is not evidence. Neither is a confident, competent-sounding voice disagreeing with you. Caution What this means for you Treat every figure a persona states as a claim to check, not a finding to act on. If a persona asserts something about your business that you did not put in the brief, it did not get it from your data. Interrupt and say so. Interruption is the primary control in the room and this is one of the things it is for. What we do about it is a set of scenarios in the decision layer's own regression suite, covering agreement under pressure, deference to a confident assertion, and a hidden cost the personas are supposed to find. The suite tests the layer that is built. It does not test a client application, because there is not one, and we are not going to describe a planned mitigation in the present tense. ## What Tingvar is not for It is not legal, medical, financial, investment, employment, insurance or safety advice. It is not a substitute for a person who is accountable for the decision. It is not a decision record, and a transcript of an argument is not a minute of a meeting. ## Verify before you act Every number a persona states is a claim to check. Every citation is a claim to check. The room is built so you can interrupt the moment you hear one, which is faster than writing it down and checking later. ## What not to put in a room Credentials and keys. Special category personal data. Other people's personal data without a lawful basis for putting it there. Anything held under an agreement that forbids third party processing. The brief reaches our model vendor with every turn, and the sub-processors page says who that is. ## Synthetic media Persona audio is generated and is labelled as generated wherever it appears on this site. Machine readable provenance marking for exported audio lands before there is an export feature, which is the cheapest time to add it. Nothing on this site is a recording of a real session. ## Emotion recognition We do not do it. The system detects whether somebody is speaking and arbitrates who holds the floor next. It does not infer mood, sentiment, stress or any other state of a person, and it does not categorise anybody by a biometric characteristic. This paragraph is the written assessment. ## EU AI Act, Article 50 | Item | Status | Evidence | | --- | --- | --- | | Article 50(1), tell the person they are talking to AI | Yes. Four synthetic voices arguing in a room is squarely the case it was written for | Stated on this page, on the marketing pages, on the screen before the room opens, and at the start of every room | | Article 50(2), mark synthetic output as machine readable | Yes, for persona audio and for anything exported | The approach is documented here. There is no export feature yet, and provenance marking lands before there is one | | Article 50(3), emotion recognition and biometric categorisation | No. The system detects that somebody is speaking and decides who holds the floor. It infers nothing about a person | The assessment is written down here, because it is a question a voice product gets asked | | Article 50(4), disclosing AI generated publications | Not to the product. It applies to us when we publish | Every synthetic asset on this site is labelled as synthetic in the asset itself and in its caption | The four transparency duties, and where each one is satisfied Scroll to see the rest ## Model changes The models behind the personas will change, and the personas will change with them. Output quality and character move when that happens. Anything that changes what the product does is dated in the build log. ## Reporting a bad output Send us the session and what went wrong. We read every one, we reply within 2 business days, and a fabrication that we can reproduce becomes a scenario in the regression suite. support@tingvar.com ## The question people ask ### Will it invent numbers about my business? A Tingvar persona once turned the misheard phrase "turns in" into a confident objection about 15 % monthly churn that had never been said. We publish this because a room whose value is pressure-testing your numbers, and which invents numbers, is worse than no room at all. The personas argue from the brief you write and from what you say, and they cannot see your accounts or anything you have not told them. If a number sounds wrong, say so out loud. Related documents - [Terms of service](/terms) - [Trust centre](/trust) - [Security](/security) --- Source: https://tingvar.com/compare Updated: 2026-09-12 24 AUG 2026 # Compare Five comparisons, every claim taken from the other product's own published pages, with the date we read it and a route to correct it. Tingvar is a live audio room where two to four AI personas argue with you, and with each other, about a plan you write beforehand. It is a debate harness. Everything below compares that against something else, and every page states where the other product is ahead of us first. ## The five comparisons - [AI Group Call The other live voice product in the set. They ship on two app stores and we do not. They publish no audio and no latency figure.](/compare/ai-group-call) - [AskVerdict The content leader, and ahead of us on almost everything except the product. Text only, so nothing in it can be interrupted.](/compare/askverdict) - [ChatGPT The real incumbent is a habit. Three prompts in one chat window cannot interrupt each other and cannot be interrupted by you.](/compare/chatgpt) - [SynthBoard They own the anti-sycophancy term outright and publish seventeen real dated sessions. Their voice mode has no published behaviour.](/compare/synthboard) - [Building it yourself You can build text multi agent debate in an afternoon. The hard part is deciding which of three willing speakers gets the floor.](/compare/build-it-yourself) ## What Tingvar is, and what it is not Tingvar is not an AI chatbot, an AI assistant, a meeting assistant, a note-taker, a summariser, a copilot or an autonomous agent platform. It is not an interrogation: nobody scores you, and nothing here is about you. The list of what it is not is longer than the list of what it is, and that is deliberate. Every label on the right describes a product that agrees with you, which is the failure this one exists to attack. | It is | It is not | Why the difference matters | | --- | --- | --- | | A live audio debate room | An AI chatbot or an AI assistant | You speak and it answers back in a room, and nothing in the room is on your side by default. | | A written brief, read before the room opens | A meeting assistant, a note-taker or a summariser | A meeting assistant sits in a meeting you were already having. This is the meeting. | | Two to four personas with opposed objectives | A copilot | A copilot works inside your framing. These attack the framing. | | One transcript you keep, at the end | An autonomous agent platform | Nothing here acts on your behalf. It argues, and you decide. | | A room you are in, in real time | An interrogation | Nobody scores you and nothing here is about you. The object of every attack is the plan. | The category, stated in both directions Scroll to see the rest ## What nobody in this category publishes Two counts, both read on 23 August 2026, and both stated with their sample size. Products publishing a voice to voice latency figure, a barge-in figure or a qualitative one: 0 of the eight properties we read MEASURED 2026-08-23, 8 readings. Products publishing a named customer, a user count, a revenue figure, a session count or a latency figure: 0 of the thirteen products we read MEASURED 2026-08-23, 13 readings. That is context for our own empty rows rather than a criticism of anyone. Outcome evidence does not exist anywhere in this category, ours included, and we are the one required by our own rules to say so. [See what we have measured](/evidence) ## Questions people ask before comparing ### Is this an AI meeting assistant? Tingvar is not an AI chatbot, an AI assistant, a meeting assistant, a note-taker, a summariser, a copilot or an autonomous agent platform. Its output is an argument rather than a record of one. A meeting assistant sits in a meeting you were already having and writes down what happened. Tingvar is the meeting, and nobody in it is on your side by default. If what you need is a transcript of a real conversation with real colleagues, this is the wrong product. ### How is Tingvar different from ChatGPT? Tingvar and ChatGPT solve different problems: ChatGPT gives you one assistant optimised to be helpful, and Tingvar gives you three personas optimised to disagree with each other. You can interrupt a Tingvar room out loud and the personas can interrupt each other, which a chat window does not do because there is nothing to interrupt. Most people who use Tingvar will keep using ChatGPT, and for most of what they do that is the right choice. The comparison page states where each one wins. ### How is Tingvar different from AI Group Call? AI Group Call runs on iPhone and Android only, and Tingvar runs in a browser with nothing to install. Both are live voice products with more than one AI participant, which makes this the closest comparison in the category. AI Group Call meters by the minute, and Tingvar prices by the 45-minute session. Tingvar publishes a measured voice-to-voice latency figure with its method and its date. The full comparison, with the crawl date on every fact, is on the comparison page. ### How is Tingvar different from AskVerdict? AskVerdict's product is text and Tingvar's is speech, so their personas cannot interrupt each other and you cannot interrupt them, because there is nothing to interrupt. Where they are ahead of us, stated here rather than left for you to find: AskVerdict ships an API, an SDK and a command line tool, and publishes seven case studies. We have none of those. What we have is a spoken room you can cut into and a measured latency figure. ### What is Tingvar bad at? Tingvar is bad at anything you want agreement or output from: it will not draft your deck, write your code, or summarise your meeting. It is a poor fit for casual brainstorming, for a decision you have already made, and for anyone who wants their instinct confirmed. It has no published customer outcomes, because zero real human sessions have taken place. The evidence page carries what we have measured and what we have not, including the rows that are empty. ## How we compare Every claim about another product on this site is a fact that product publishes about itself, read from its own pages on the date below. Where a product publishes nothing on a point, the cell reads not published, because an absent published fact is not an absent capability. Our own weaknesses are in the same table, in the same column style. Compared on 23 August 2026 ### Sources | Claim | Read on their own pages at | Read on | | --- | --- | --- | | No crawled competitor publishes a latency figure of any kind | A full read of eight competitor properties, roughly 530 URLs | 23 August 2026 | | None of thirteen products publishes a named customer, a user count, a revenue figure, a session count or a latency figure | A read of every reachable page across thirteen products | 23 August 2026 | | AI Group Call ships on iOS and Android and states no web client | aigroupcall.app, and the JSON-LD on their own homepage | 23 August 2026 | | AskVerdict has no page mentioning voice across 195 pages | www.askverdict.ai, full corpus read | 23 August 2026 | | SynthBoard publishes seventeen dated shared sessions | www.synthboard.ai shared sessions index | 23 August 2026 | Every claim about each product compared on this page, with its source and the date it was read Scroll to see the rest ### Right of reply If you work at one of these companies and something here is wrong, email corrections@tingvar.com. We correct it within 1 business day and log the correction publicly. [corrections@tingvar.com](mailto:corrections@tingvar.com) ### Correction log No correction has been requested or issued. This log is dated and it stays here whether or not it has entries. --- Source: https://tingvar.com/compare/ai-group-call Updated: 2026-09-12 24 AUG 2026 # Tingvar vs AI Group Call Both are live audio rooms with several AI voices in them. Tingvar runs in a browser and publishes what its latency was measured at. AI Group Call runs on iOS and Android and publishes neither a web client nor a figure. ## Where they are ahead of us They shipped. AI Group Call is on the App Store and on Google Play: iOS version 1.0 released 22 August 2026. Source: Their App Store listing. Read on 23 August 2026. People have installed it and used it. Our client application is not started, we are in no store, and zero real human sessions have taken place. That is the largest difference between the two products and it is not in our favour. They are also the other live voice product in the crawled set. Six AI participants cast as a host, a sceptic, a strategist and others, read from aigroupcall.app homepage subhead on 23 August 2026, with an editable cast, a live transcript and interruption their own copy describes plainly. That is a larger cast than we open a room with, and a reader choosing on the size of the room should count it in their favour. Every other product in the crawled set is text. If what you want is a voice room on a phone this week, they have one and we do not. Their unit of account is finer than ours. Read from aigroupcall.app pricing section on 23 August 2026: Pricing metered in minutes per month, with no purchase control in the pricing section. The first half of that is ahead of us. They sell minutes and we sell a session, so a person who wants to be charged for the time they actually spend has a unit there and none here. The second half is not an advantage to them and we will not dress it as one. There is no price comparison anywhere on this page, because putting the two side by side would need us to model their meter, and a comparison resting on our model of somebody else's meter is not something you could check. Add those up before you read the rest of the page. They have a released application on two stores, a larger cast, a finer unit of account and people who have used the product. We have a browser, a measurement run and a set of published blind spots. Nothing below softens that. ## Where the two products actually diverge Two things, and both are checkable from their pages rather than from ours. The first is the browser. Their structured data lists Android and iOS as the operating systems the application runs on, and every call to action on their site ends in an app store, so a person at a desk has nothing to open. Tingvar's first surface is the browser and the mobile applications are the port. The second is that we publish a number. Voice to voice, from you finishing a sentence to the first persona's audio, has a median of 1471 ms MEASURED 2026-08-19, 122 readings. The run behind it was 122 readings over 200 utterances against Vertex AI in us-central1 and a real media server, taken on 19 August 2026. There is nothing to compare that against. Of the competitor properties read on 23 August 2026, the number publishing a latency figure of any kind is 0 of the eight properties we read MEASURED 2026-08-23, 8 readings. That is a statement about what is published, not about who is faster. Barge-in is speaking over a persona so that it stops, and your voice takes the floor from whoever is holding it. Human speech preempts everything in the room. That is an invariant of the design rather than a setting you switch on, and it is why speaking is how you steer a session at all. Speaking is also how you change the subject, ask a persona to defend a claim, or stop a line of argument that is going nowhere. Barge-in is not a raised hand, not a queued turn and not a pause button. The client leg of barge-in is not yet measured, and we publish measurements rather than targets. ## Row by row | | Tingvar | AI Group Call | | --- | --- | --- | | Modality | Live audio room | Live audio call | | Runs in a browser | Yes, and it is the first surface | No. Their own structured data lists Android and iOS, and every call to action ends in an app store | | iOS app | Not built | Yes, version 1.0, released 22 August 2026 | | Android app | Not built | Yes | | Voices in a room | Two to four, three by default | Six | | What separates the voices | What each optimises for, plus a published blind spot for each | A cast of roles: a host, a sceptic, a strategist and others | | Published blind spots | Yes, one per persona, on /personas | Not published | | Interruption | Human speech takes the floor from whoever holds it | Claimed in body copy beside a hand built mockup. No published behaviour | | Published latency figure | Median 1471 ms, 95th percentile 1986 ms, from 122 readings in us-central1 on 19 August 2026 | Not published | | Published audio of the product | Four persona voice samples, labelled as scripted. No session recording | Not published. Both homepage demonstrations are hand built HTML with fixed strings | | A written brief before the room opens | Yes, and the personas have read it on arrival | Not published. Their flow starts from a typed goal | | Unit of account | One 45 minute session | Minutes per month | | Real human sessions we have published | None. Zero real human sessions have taken place | Not applicable. They have shipped and we have not | | Public API | Not at launch | Not published | Tingvar and AI Group Call on facts each publishes about itself, read on 23 August 2026. Our own gaps are in the same column style as theirs. Scroll to see the rest ## When to use each ### Reach for AI Group Call - You want a voice room on a phone, today. - You want six voices rather than a smaller cast. - You want to start from a typed goal rather than write a brief first. - You are choosing on what has shipped rather than on what is published. ### Reach for Tingvar - You are at a desk with a written plan you want attacked. - You want to know what each voice is optimising for, and what it is wrong about. - You want the latency figure and the method behind it. ## Questions about this comparison ### How is Tingvar different from AI Group Call? AI Group Call runs on iPhone and Android only, and Tingvar runs in a browser with nothing to install. Both are live voice products with more than one AI participant, which makes this the closest comparison in the category. AI Group Call meters by the minute, and Tingvar prices by the 45-minute session. Tingvar publishes a measured voice-to-voice latency figure with its method and its date. The full comparison, with the crawl date on every fact, is on the comparison page. ## How we compare Every claim about another product on this site is a fact that product publishes about itself, read from its own pages on the date below. Where a product publishes nothing on a point, the cell reads not published, because an absent published fact is not an absent capability. Our own weaknesses are in the same table, in the same column style. Compared on 23 August 2026 ### Sources | Claim | Read on their own pages at | Read on | | --- | --- | --- | | No web client, and structured data listing Android and iOS as the operating systems | aigroupcall.app homepage and its own JSON-LD | 23 August 2026 | | Six AI participants cast as a host, a sceptic, a strategist and others | aigroupcall.app homepage subhead | 23 August 2026 | | Barge-in claimed in body copy beside a hand built mockup, with no audio published anywhere | aigroupcall.app homepage caption and a full read of all five pages | 23 August 2026 | | Both homepage demonstrations are hand built HTML with fixed strings | The shipped markup of aigroupcall.app | 23 August 2026 | | iOS version 1.0 released 22 August 2026 | Their App Store listing | 23 August 2026 | | Pricing metered in minutes per month, with no purchase control in the pricing section | aigroupcall.app pricing section | 23 August 2026 | | No public API, no MCP server and no browser extension | A full read of all five pages of aigroupcall.app | 23 August 2026 | Every claim about AI Group Call on this page, with its source and the date it was read Scroll to see the rest ### Right of reply If you work at one of these companies and something here is wrong, email corrections@tingvar.com. We correct it within 1 business day and log the correction publicly. [corrections@tingvar.com](mailto:corrections@tingvar.com) ### Correction log No correction has been requested or issued. This log is dated and it stays here whether or not it has entries. ## Hear the difference for yourself There is no migration offer here and no switching discount. Most people evaluating both should try both. [Hear the four voices](/personas) Start a session --- Source: https://tingvar.com/compare/askverdict Updated: 2026-09-12 24 AUG 2026 # Tingvar vs AskVerdict Both put several AI personas against one decision. AskVerdict does it in writing, so the personas take turns. Tingvar does it out loud, so they cut each other off and so can you. ## Where they are ahead of us On almost everything except the product. Read from www.askverdict.ai/sitemap.xml and www.askverdict.ai/docs on 23 August 2026: 169 URLs declared in their own sitemap, a 58 page documentation tree, a REST interface, a TypeScript library, a command line tool and streaming schemas. It also has a trust surface with a data processing agreement and a sub-processor list, a status page, a changelog, and the option to bring your own model key. We will not match that content surface at launch and we are not going to pretend otherwise. One of their pages is better than anything we had before we copied it. A calibration page that explains a method and reports insufficient data to plot the curve, read from www.askverdict.ai/calibration on 23 August 2026. That is a page that concedes something, and it is the reason a reader believes the rest of the site. Our evidence page is built on the same principle and it exists because of theirs. Their writing also already ranks on the argument we are making. If you are choosing on how much has been published about a category, they win that comparison and it is not close. ## One material difference, and it is the medium AskVerdict is text. A read of 195 of their pages found no page about voice, and every use of the word on the site is a metaphor. That is not a criticism of the writing. It is a fact about what the product can do, and it has one consequence that matters: in text there is nothing in progress, so nothing can be cut off. Their personas rebut each other in completed written turns and you read the result afterwards. In a room, one persona holds the floor at a time and your voice takes it from whoever has it, mid sentence. That is the whole of what the second product buys you over the first, and if it is not worth anything to you then AskVerdict is the better choice today. ## Row by row | | Tingvar | AskVerdict | | --- | --- | --- | | Modality | Live audio room | Text. A read of 195 pages found no page about voice | | Can you interrupt it | Yes. Speaking takes the floor from whoever holds it | There is nothing in progress to interrupt | | Can the agents interrupt each other | Yes, and the room decides who holds the floor | No. Their agents rebut each other in written turns | | What separates the agents | What each optimises for, plus a published blind spot for each | Six personas, described by role | | Published blind spots | Yes, one per persona, on /personas | Not published. Their own /personas returns 404 | | Published latency figure | Median 1471 ms, 95th percentile 1986 ms, from 122 readings in us-central1 on 19 August 2026 | Not published | | Public API | Not at launch | Yes. REST, a TypeScript library, a command line tool and streaming, across 58 documented pages | | Case studies | None. Zero real human sessions have taken place | Seven, and none of them names a company or a person | | Trust surface | A trust centre, a sub-processor list naming the model vendor, and a pre-signed data processing agreement | A data processing agreement and a sub-processor list of five vendors, none of them a model vendor | | Comparison values without JavaScript | Server rendered text in every cell | Their feature matrix values and fifteen pricing answers are rendered by script and are absent without it | | Hero positioning | One statement, and no test of it at launch | Four hero variants ship at once, with a rotating line that matches none of them | | An empty metric published as empty | Six defined metrics with targets and no values, on /evidence | Their calibration page reports insufficient data to plot its curve | Tingvar and AskVerdict on facts each publishes about itself, read on 23 August 2026. Our own gaps are in the same column style as theirs. Scroll to see the rest ## When to use each ### Reach for AskVerdict - You want to run this from your own code, today, against a documented interface. - You want a written record to circulate rather than an argument to sit through. - Your procurement process needs a trust surface that already exists. - You want to read the output later rather than be in the room now. ### Reach for Tingvar - You want to be interrupted, and to interrupt. - You want to hear where two advisers split, at the moment they split. - You want each persona's blind spot published rather than inferred. ## Questions about this comparison ### How is Tingvar different from AskVerdict? AskVerdict's product is text and Tingvar's is speech, so their personas cannot interrupt each other and you cannot interrupt them, because there is nothing to interrupt. Where they are ahead of us, stated here rather than left for you to find: AskVerdict ships an API, an SDK and a command line tool, and publishes seven case studies. We have none of those. What we have is a spoken room you can cut into and a measured latency figure. ## How we compare Every claim about another product on this site is a fact that product publishes about itself, read from its own pages on the date below. Where a product publishes nothing on a point, the cell reads not published, because an absent published fact is not an absent capability. Our own weaknesses are in the same table, in the same column style. Compared on 23 August 2026 ### Sources | Claim | Read on their own pages at | Read on | | --- | --- | --- | | Text only, with no page about voice across 195 pages | www.askverdict.ai, full corpus read | 23 August 2026 | | 169 URLs declared in their own sitemap, a 58 page documentation tree, a REST interface, a TypeScript library, a command line tool and streaming schemas | www.askverdict.ai/sitemap.xml and www.askverdict.ai/docs | 23 August 2026 | | Seven case studies naming no company and no person | www.askverdict.ai case study pages | 23 August 2026 | | A calibration page that explains a method and reports insufficient data to plot the curve | www.askverdict.ai/calibration | 23 August 2026 | | Their own /personas, /faq, /support, /trust, /enterprise and /refund-policy return 404 | Direct requests to each path on www.askverdict.ai | 23 August 2026 | | A sub-processor list of five vendors, none of them a model vendor | www.askverdict.ai sub-processor page | 23 August 2026 | | Feature matrix values and fifteen pricing answers rendered by script and absent without it | www.askverdict.ai/pricing, fetched with scripting disabled | 23 August 2026 | | Four hero variants shipping at once, with a rotating line matching none of them | The shipped JavaScript bundle of www.askverdict.ai | 23 August 2026 | Every claim about AskVerdict on this page, with its source and the date it was read Scroll to see the rest ### Right of reply If you work at one of these companies and something here is wrong, email corrections@tingvar.com. We correct it within 1 business day and log the correction publicly. [corrections@tingvar.com](mailto:corrections@tingvar.com) ### Correction log - 26 August 2026 The documentation tree was published at two different sizes on this page, one in the text and one in the source ledger below. Both now read 58 pages, which is the number of pages under their own /docs on the date beside every claim here. ## The part you cannot read about Whether being argued at out loud is worth anything is not something a comparison table can settle. [Hear the four voices](/personas) Start a session --- Source: https://tingvar.com/compare/chatgpt Updated: 2026-09-12 24 AUG 2026 # Tingvar vs ChatGPT The thing most people compare us against is not a product. It is a habit: open one chat window and ask the same question three ways. ## Where they are ahead of us It is already open, it costs you nothing to try, it needs no written brief, and it answers in about a second. It does hundreds of things well and this does one thing. Most people who read this page will keep paying for a general assistant afterwards, and they should. Almost everything on our side of this comparison is either unbuilt or unused. Our client application is not started, we are in no app store, we publish no interface your own code can call at launch, and zero real human sessions have taken place. The habit needs none of that. It needs a window you already have open, and it asks you for nothing before it answers. A reader choosing on what exists today rather than on what is published should choose the window, and this page is not written to talk them out of it. The one thing we say about the habit is also the weakest sourced claim on the site, and it belongs here rather than in the small print. One model answers in the order you ask, and each answer completes before the next begins. Source: Reproducible in any chat window. This page makes no claim about a published feature. Read on 24 August 2026. Every other comparison page cites a page the other product published; this one cites a thing you can reproduce in your own window in a minute. That is a lower standard of evidence and it is the standard we have. There is no switching offer on this page and no migration discount. If the way you already work is producing decisions you are happy with, the honest recommendation is to carry on. ## Why the column does not say ChatGPT Every claim this site makes about another product is a fact that product publishes about itself, read from its own pages on a stated date. We have not crawled OpenAI's site, so we make no claim about a ChatGPT feature, a ChatGPT price or a ChatGPT plan, and the right hand column does not pretend to. What it holds instead is a property of a single chat thread, which you can reproduce in your own window in a minute. That is a weaker claim than a feature grid and it is one we can stand behind. ## Three prompts in one window is a different machine Ask one assistant to be the finance view, then the engineering view, then the growth view, and you get three answers written by one reader of your plan, in the order you asked for them. None of them heard the others say it. None of them can object while it is being said. The sequence is yours to set, which means the disagreement is yours to arrange, which means it tells you what you already suspected. In a room, the second voice objects to the first one while the first one is still talking, and the objection is aimed at the sentence rather than at a summary of it. When two of them reach opposite conclusions from the same paragraph, the assumption they price differently is the one carrying your plan's weight. That is a location, and one reader cannot give you it however many times you ask. [Read the thesis](/why-disagreement) ## Row by row | | Tingvar | Three prompts in one chat window | | --- | --- | --- | | How many voices | Two to four, three by default, in one room | One model, answering in the order you ask | | Who decides who speaks | The room, once per turn, from every persona's bid | You, by typing the next prompt | | Can they interrupt each other | Yes, and the interruption is audible | No. Each answer is complete before the next begins | | Can you interrupt | Yes. Speaking stops whoever is talking, mid sentence | You can stop a response and start again. There is no sentence in progress to cut into | | How long a critical instruction holds | It is not an instruction. Each persona is optimising for a different thing for the whole session | Be critical relaxes back toward agreement after two or three turns | | What separates the perspectives | What each optimises for, plus a published blind spot for each | Whichever roles you describe in the prompt, read by one model | | Where the argument lives | Between the personas, and you listen to it | Between you and the assistant | | Cost to try | 149 USD a month, refunded in full if you end your first session inside 5 minutes | You already have it open | | Breadth of what it does | One thing: it argues about a written plan | Almost everything most people use an assistant for | | Real human sessions we have published | None. Zero real human sessions have taken place | Not applicable | Tingvar against the habit of asking one assistant three times in one thread. The right hand column describes a way of working, not a published feature list. Scroll to see the rest Not yet produced A static figure putting the three arrangements side by side. It is producible today and it has not been drawn. Both pages carry the same three states as a table, so the gap costs a reader the picture. Blocked on design production 2026-08-26 ## When to use each ### Keep using the chat window - You want an answer rather than an argument. - The question has a right answer and you want it in a second. - You are drafting, summarising, coding or translating. - You do not have a written plan yet. ### Open a room - You have written the plan and you suspect you are too close to it. - You want to hear the objection while you can still answer it. - You want to know which assumption two competent readers price differently. ## Questions about this comparison ### How is Tingvar different from ChatGPT? Tingvar and ChatGPT solve different problems: ChatGPT gives you one assistant optimised to be helpful, and Tingvar gives you three personas optimised to disagree with each other. You can interrupt a Tingvar room out loud and the personas can interrupt each other, which a chat window does not do because there is nothing to interrupt. Most people who use Tingvar will keep using ChatGPT, and for most of what they do that is the right choice. The comparison page states where each one wins. ### How is this different from three ChatGPT tabs? Three ChatGPT tabs give you three answers, and a Tingvar room gives you an argument, because separate tabs cannot hear each other. Nothing in tab two knows what tab one said unless you paste it there yourself, and by then you have chosen which criticism survives. Inside one chat window the instruction to be critical decays after two or three turns. In a room, one persona attacks another persona's reasoning while you listen, and you can cut in the moment either of them goes wrong. ## How we compare Every claim about another product on this site is a fact that product publishes about itself, read from its own pages on the date below. Where a product publishes nothing on a point, the cell reads not published, because an absent published fact is not an absent capability. Our own weaknesses are in the same table, in the same column style. Compared on 23 August 2026 ### Sources | Claim | Read on their own pages at | Read on | | --- | --- | --- | | One model answers in the order you ask, and each answer completes before the next begins | Reproducible in any chat window. This page makes no claim about a published feature | 24 August 2026 | | A critical instruction relaxes back toward agreement after two or three turns | Our own testing, and the argument set out on /why-disagreement | 24 August 2026 | | Post-training optimises assistants for helpfulness and for user satisfaction | A general property of instruction tuned assistants, argued rather than measured, on /why-disagreement | 24 August 2026 | | Tingvar median voice to voice latency of 1471 ms from 122 readings in us-central1 | Our own measurement run, published in full on /evidence | 19 August 2026 | Every claim about a single chat thread on this page, with its source and the date it was read Scroll to see the rest ### Right of reply If you work at one of these companies and something here is wrong, email corrections@tingvar.com. We correct it within 1 business day and log the correction publicly. [corrections@tingvar.com](mailto:corrections@tingvar.com) ### Correction log No correction has been requested or issued. This log is dated and it stays here whether or not it has entries. ## Bring the plan you already asked about Professional is 149 USD a month, and if you end your first session inside 5 minutes we refund it in full. Start a session --- Source: https://tingvar.com/trust/your-data Updated: 2026-09-09 # Your data A Tingvar session produces a transcript. That transcript is a record of your company's unresolved strategic weaknesses, spoken aloud and attributed. Read this before you run one. Version 1.0.0 Effective 2026-08-24 Last updated 2026-08-24 In short Applies to: Every buyer, before their first session - A session transcript is a record of your unresolved strategic weaknesses. - This page states where it goes, who can reach it and how to destroy it. - Read it before your first session rather than after it. ## What a session actually produces This is the inventory a reviewer builds for themselves during due diligence. Publishing it first is both honest and faster for everybody. | Artefact | What it contains | Where it is created | Sensitivity | | --- | --- | --- | --- | | Your microphone audio | Your voice, saying your company's problems out loud, and anything else audible in your room | Your browser, the media server, then the model vendor | High | | The transcript of what you said | The text of everything you said in the room | The speech vendor, in the United States | High | | The brief | The proposal you want attacked. It becomes part of every persona's instructions | Typed on this site, then sent to the model vendor with every turn | Highest | | Persona audio | Synthetic speech, generated from your brief and what you said | The model vendor, the media server, then your browser | Medium | | The attributed transcript | Who said what, which assumption was attacked, and where the personas disagreed | Our systems | Highest | | Session metadata | Duration, which personas were in the room, floor share, interruption counts, cost | Our systems | Low | | Derived quality measures | How often the personas agreed with you, how often they disagreed with each other | Our systems | Low | | The answer to the closing question | What you now believe that you did not believe an hour ago | Our systems | Medium | Everything one session creates, and where it is created Scroll to see the rest ## How long each piece is kept Every row carries a number. An unbounded retention statement is banned from this site by our own commitment C20, and the build fails on the phrase it usually hides behind. | Item | How long | How to delete it | | --- | --- | --- | | Session audio, yours and the personas' | Not retained. Processed while the room is live | Nothing to delete, because nothing is written | | Transcripts and the brief | Kept until you delete them, then gone within 30 days including backups | Delete a session yourself, or delete the account | | Session metadata, such as duration and floor share | 24 months, then aggregated | Aggregated records carry no identifier | | The one question a session ends on | Destroyed with the session, so within 30 days of you deleting it | Deleted with the session | | Server logs | 90 days | Purged automatically | | Backups | Purged within 30 days of the source being deleted | Automatic, and this is the number questionnaires ask for | | Billing records | 7 years | Tax law requires them, so they survive a deletion request | | Consent records | 6 years after the consent is withdrawn | Kept to evidence that the choice was honoured | The retention schedule, with a number on every row Scroll to see the rest ## Where it is processed Your session audio and your transcripts are processed in the United States, in Google Cloud's us-central1 region. There is no EU only processing region for this configuration. For customers in the UK and the EEA that is a transfer out of your region, and we rely on Standard Contractual Clauses and the safeguards in our data processing agreement. Caution If your procurement requires EU residency Today the honest answer is that we cannot meet it. We would rather you knew that before you bought than after. When the position changes we will publish it here with a date, and we will not promise one in advance. Not yet produced A figure showing where your words go and where they stop. It has to match the sub processor list exactly, and that list is not final, so drawing it now would publish a boundary we might move. Blocked on the sub processor list being final 2026-08-26 ## What is not built The compliance layer is not built. That is the second half of the honest answer about your data, and a reviewer should read it here rather than find it three weeks into a procurement. Named plainly, this is what is absent today, in the same words the capability table on the enterprise page uses. - No single sign on. - No directory provisioning. - No audit log and no audit export. - No tenancy isolation. Sessions are isolated per account, not per tenant. - No retention override. Retention is the published default for every account. - No role model beyond one role. Every account holder can do everything their own account can do. - No service level agreement. Not planned. There is no operating history to write one from. Each of these is stated as not built, not as coming. Where an item sits on the roadmap it sits there in the future tense, with the state it is in and no delivery date, because a date we cannot keep is worth less to you than the word no. If one of them is a hard requirement, the honest answer today is that we cannot meet it. The evidence is published in two tables rather than asserted here. The [capability table on the enterprise page](/enterprise) carries a yes or a no on every row, and the [known gaps on the security page](/security) name what we do not have and why. ## Who touches it Every vendor that touches customer data is named, including the model vendor, which is the row most vendor lists leave out. The list carries the data category each one receives and the jurisdiction it operates in. [The full sub-processor list](/subprocessors) ## The questions this page exists to answer ### What happens to my recording? Tingvar does not store your session audio: it is processed while the room is live and no recording of it is kept. What persists is the transcript, the text of what you said and what each persona said, attributed line by line. Only the account that ran the session can read that transcript, and nobody at Tingvar reads one without your written request on a support case. You can delete any session yourself, and deletions are purged from backups within 30 days. The retention schedule is published at tingvar.com/trust/your-data. ### Who can hear my strategic weaknesses? Nobody else joins a Tingvar room: the participants are you and the AI personas, and the audio is not stored for anyone to play back later. The transcript is the part that persists, and it reaches our model vendor as part of the personas' context. Every vendor that touches customer data is named on the sub-processors page, including the model vendor. Who at Tingvar can read a transcript, and in what circumstances, is stated at tingvar.com/trust/your-data. ### Is my data used to train AI models? No, Tingvar does not use your brief, your transcript or your audio to train models. Google, which runs the model, states that customer data submitted to Vertex AI is not used to train or fine-tune its models without permission, that cached content is held for up to 24 hours in the data centre serving the request, and that prompts may be logged for abuse detection unless an exception is agreed. It is linked from our sub-processors page so you can check it. ### Where is my data stored? Your Tingvar session audio and transcripts are processed in the United States, in Google Cloud's us-central1 region, and we do not currently offer an EU-only processing region. For customers in the UK and the EEA, that means a transfer out of your region. We rely on Standard Contractual Clauses and the safeguards in our DPA. If your procurement requires EU data residency, the honest answer today is that we cannot meet it, and we would rather you knew that before you bought than after. ### How do I delete a session? You delete a Tingvar session yourself from the session list, and deleting your account destroys every session in it. Deletion is self-service and does not go through support. Anything you delete is purged from backups within 30 days, and we publish that number because an unbounded backup window is what fails a security questionnaire. Billing records survive deletion for seven years, because tax law requires it, and we say so rather than implying everything vanishes. ### I am going to describe things about my company that are not public. Should I? A Tingvar session is a spoken record of what is wrong with your plan, which makes it one of the most sensitive artefacts your company could produce. So the honest advice is to read our data page before your first session rather than after it. What we control is where it goes, who can reach it and how fast you can destroy it, and all three are published. Names, unreleased figures and anything under an agreement with somebody else are worth redacting from the brief. ### Do you have SOC 2 or ISO 27001? No, Tingvar holds no SOC 2 report and no ISO 27001 certificate, and we will not claim a stage of either one that we have not reached. No product in the set we crawled on 23 August 2026 publishes one either. What we do have is published: the controls on the security page, a pre-signed DPA you can read in full without a form, and every sub-processor named including the model vendor. We answer a security questionnaire in five business days. ## What we would redact if it were our plan Names of individuals, unreleased figures, and anything held under an agreement with somebody else. The personas argue about the shape of a plan, and a shape survives redaction better than most people expect. Related documents - [Privacy notice](/privacy) - [Sub-processors](/subprocessors) - [Security](/security) - [Data processing agreement](/dpa) --- Source: https://tingvar.com/subprocessors Updated: 2026-09-08 # Sub-processors Every vendor that touches customer data, what it receives, and where it operates. The model vendor is on the list, which is the row most vendor lists leave out. Version 1.0.0 Effective 2026-08-24 Last updated 2026-08-24 In short Applies to: Procurement and security reviewers - The model vendor is named, which is the row most vendor lists leave out. - The underlying cloud is named for any vendor reselling infrastructure. - Changes carry 30 days of notice before processing begins. ## The list A row marked as not chosen is a decision we have not taken. It is on the list so that a reviewer can see the shape of the finished list rather than discovering a new vendor after signature. | Vendor | What it does | Data it receives | Where | Status | | --- | --- | --- | --- | --- | | Google Cloud, Vertex AI | Runs the personas. Receives the brief, the session audio and the transcript | Your audio, your transcript, the brief, persona speech | us-central1, United States | Confirmed | | Google Cloud, Speech-to-Text | Transcribes what you say so the personas can read it | Your audio, your transcript | us-central1, United States | Confirmed | | Media server | Carries audio between your browser and the personas | Your audio and persona audio, in transit | Decided with the vendor | Vendor not chosen yet | | Paddle, merchant of record | Takes subscription payments as the seller of record, and issues the invoice. No card number reaches us | Billing identifiers and transaction metadata | Decided with the vendor | Vendor not chosen yet | | Application hosting | Serves this website and the application | Request metadata, account data, transcripts, the brief | Decided with the vendor | Vendor not chosen yet | | Brevo, transactional email | Sends verification, receipts, billing notices and erasure certificates | Your email address and the content of the message | Decided with the vendor | Vendor not chosen yet | | WorkOS, enterprise federation | Carries single sign-on and directory sync for a company that signs in through its own identity provider | Your email address, your name, and the group memberships your employer syncs | Decided with the vendor | Vendor not chosen yet | | Error monitoring | Reports faults so they can be fixed | Stack traces. No request bodies | Decided with the vendor | Vendor not chosen yet | Sub-processors, effective 24 August 2026 Scroll to see the rest Total vendors that will touch customer data: 8 ## What the model vendor does with your session Audio and text sent to Vertex AI are processed to generate the personas' responses. Google states that customer data submitted to Vertex AI is not used to train or fine-tune its models without permission, that cached content is held for up to 24 hours in the data centre serving the request, and that prompts may be logged for abuse detection unless an exception is agreed. [Google, Vertex AI data governance opens in a new tab](https://cloud.google.com/vertex-ai/generative-ai/docs/data-governance) That is Google's commitment and not ours. We do not restate a vendor's promise as though we made it. Ours is the sentence above it: we do not use your sessions to train anything. ## Sub-sub-processors Where a vendor is itself reselling infrastructure, the underlying cloud is named in its row. A list that stops at the reseller tells a reviewer nothing, and it is the omission our own extraction criticises in the vendor lists we read. ## Changes Note Thirty days of notice A new sub-processor gets 30 days of notice before it begins processing customer data. The effective date at the top of this page is how you tell how current the list is. To be told when it changes, write to the address below with the word subprocessors in the subject. There is no form and no marketing list behind it. legal@tingvar.com ## Related questions ### Who can hear my strategic weaknesses? Nobody else joins a Tingvar room: the participants are you and the AI personas, and the audio is not stored for anyone to play back later. The transcript is the part that persists, and it reaches our model vendor as part of the personas' context. Every vendor that touches customer data is named on the sub-processors page, including the model vendor. Who at Tingvar can read a transcript, and in what circumstances, is stated at tingvar.com/trust/your-data. ### Is my data used to train AI models? No, Tingvar does not use your brief, your transcript or your audio to train models. Google, which runs the model, states that customer data submitted to Vertex AI is not used to train or fine-tune its models without permission, that cached content is held for up to 24 hours in the data centre serving the request, and that prompts may be logged for abuse detection unless an exception is agreed. It is linked from our sub-processors page so you can check it. ### Who are your sub-processors? Tingvar publishes every vendor that touches customer data at tingvar.com/subprocessors, including the model vendor, with the jurisdiction each one operates in. The list names the underlying cloud for any vendor that is itself reselling infrastructure, because a list that stops at the reseller tells a reviewer nothing. We give 30 days' notice before a new sub-processor begins processing customer data, and you can subscribe to changes. The page carries an effective date, so you can tell how current it is. Related documents - [Privacy notice](/privacy) - [Security](/security) - [Data processing agreement](/dpa) --- Source: https://tingvar.com/apps Updated: 2026-09-12 24 AUG 2026 # Where Tingvar runs Tingvar is a live audio room. AI personas, each optimising for something different, read the plan you write beforehand and then argue about it with you and with each other. You interrupt whenever you like. Measured voice-to-voice latency is p50 1,471 ms. Zero real human sessions have taken place. Six surfaces are planned and none of them is finished. This page carries the state of each one, the date that state was confirmed, and the page where the detail lives. It is the shortest honest answer to whether this will run on the thing in front of you: not yet, on any of them. The method behind that latency figure, the sample it was taken from and the date it was measured are on [the evidence page](/evidence). Start a session Read how it works ## The six, as a table | Surface | State today | Shipping date | Where it would be listed | | --- | --- | --- | --- | | Web app | In development, since 24 Aug 2026 | No date yet | Nowhere. It is not a store surface | | iOS | In development, since 24 Aug 2026 | No date yet | App Store | | Android | In development, since 24 Aug 2026 | No date yet | Google Play | | Windows | Not planned, since 27 Aug 2026 | No date yet | Microsoft Store | | Browser extension | In development, since 24 Aug 2026 | No date yet | Chrome Web Store, Edge Add-ons, Firefox Browser Add-ons and Safari Extensions | | MCP server | In development, since 24 Aug 2026 | No date yet | Nowhere. It is not a store surface | Every surface, and what is true about each one today Scroll to see the rest The state of the client PENDING The Tingvar client is being built. What this site shows you is the design of it and the measured behaviour of the layer underneath it. There are no product screenshots on this page, because there is no product screen to photograph yet, and a render dressed as a screenshot would be the first thing we asked you not to trust us about. ## Store listings Every store any surface would be listed in is below, and not one of those listings exists. A badge here would link to a page that is not there, so there is no badge on this page. Each line says which listing is missing. App Store PENDING No App Store listing exists yet, so there is no badge here and nothing to install. Google Play PENDING No Google Play listing exists yet, so there is no badge here and nothing to install. Microsoft Store PENDING No Microsoft Store listing exists yet, so there is no badge here and nothing to install. Chrome Web Store PENDING No Chrome Web Store listing exists yet, so there is no badge here and nothing to install. Edge Add-ons PENDING No Edge Add-ons listing exists yet, so there is no badge here and nothing to install. Firefox Browser Add-ons PENDING No Firefox Browser Add-ons listing exists yet, so there is no badge here and nothing to install. Safari Extensions PENDING No Safari Extensions listing exists yet, so there is no badge here and nothing to install. ## One page each Each surface publishes its own state, its own requirements and the questions particular to it. A phone page that borrowed the browser page's answers would be wrong about battery, about interruptions and about what happens when the operating system takes the microphone away. - Web app In development 24 Aug 2026 The architecture assumes a browser. WebRTC and the Web Audio API are native there, so this is the surface every other one is a port of. [Read the Web app page](/web) - iOS In development 24 Aug 2026 LiveKit ships a Swift SDK and the platform carries a spatial audio engine. No App Store listing exists, so no badge renders. [Read the iOS page](/ios) - Android In development 24 Aug 2026 LiveKit ships a Kotlin SDK. No Google Play listing exists, so no badge renders. [Read the Android page](/android) - Windows Not planned 27 Aug 2026 MSIX packaging. Nobody in the crawled competitive set ships a desktop application, so there is no listing to compare against and none of ours. [Read the Windows page](/windows) - Browser extension In development 24 Aug 2026 Four stores, four review queues, four listings. All four are shown at once rather than detected, because a detected single link hides three answers. [Read the Browser extension page](/extension) - MCP server In development 24 Aug 2026 SynthBoard already ships an MCP server with six automation adapters. We are second here, so the page competes on the connection details and on a published latency figure rather than on being first. [Read the MCP server page](/mcp) 19 AUG 2026 ## What is measured, and what is not Voice to voice 1471 ms Median, on the bench in us-central1, with a real media server MEASURED 2026-08-19, 122 readings How we measured it Real human sessions 0 sessions Have taken place. Every measured turn so far used pre-rendered audio MEASURED 2026-08-23, 1 reading What happens next IDLE ## Nobody has used this yet Zero real human sessions have taken place. Every measured turn to date used pre-rendered audio clips rather than a live human. What is built and verified is the decision layer: the turn-taking policy, the bid scoring and the configuration guards, with 619 automated tests passing and 81.78 % combined coverage. Everything there says the system works. Nothing there says anyone wants it, and being first is a real risk you are taking. Both figures above sit on [the evidence page](/evidence) with the method and the sample size. Start a session Read about the browser surface --- Source: https://tingvar.com/mcp Updated: 2026-09-12 24 AUG 2026 # Convene a room from inside your agent The proposal is already in the agent context. An MCP server lets that agent convene a room from where the proposal is, hand it to three personas who disagree with each other, and get the transcript back. MCP is a request and response protocol. An MCP client cannot hold a live microphone. That is why the tool returns a join URL rather than a debate. What exists today In development 24 Aug 2026 The server is being built, no address has been issued and no key can be created today. p50 1,471 ms, p95 1,986 ms, p99 2,156 ms, voice to voice, from the end of a human utterance to the first sample of persona audio at the client. Method: 122 readings over 200 utterances against Vertex AI in us-central1 and a real SFU, taken on 19 Aug 2026. The figure describes the room the join URL opens and not the MCP round trip, and [the evidence page](/evidence) carries the same run broken into its stages. The MCP tool call round trip is NOT MEASURED. What would measure it is a timed tools/call against the live endpoint from each supported client, at least 100 tool calls per client, reported at p50 and p95 with the client named and dated. Start a session Read the MCP docs In the browser you already have. ## Be told when it ships One email, on the day this surface ships, and nothing else. Off unless you tick it, and one click to stop. 19 AUG 2026 ## What there is to show, and what is measured ### What there is to show, and what there is not There is no recording of a real call, because no call has been made. What stands in its place is the shape the first one has: the command that adds the server, and one tool call with the argument it takes. It has never been run, because there is no endpoint to run it against, so the date on the label is the date the shape was last confirmed and not a date it executed. open\_room is a proposed tool name and the tool list is not frozen. INTENDED SHAPE. The server is not live. 2026-09-09 ``` # Add the server to a client. One command, the transport and the address. claude mcp add --transport http tingvar https:///mcp # One tool call, in the shape a client sends it. { "method": "tools/call", "params": { "name": "open_room", "arguments": { "brief": "Split the shared database behind billing, invoices first. Say what breaks." } } } # What comes back is a join URL and the time it expires. A person opens it # and speaks in the room. The transcript is read back when the room ends. ``` Scroll to see the rest Copy The call spends nothing on its own. It returns a join URL, and a session is spent when a person opens that URL and the room starts, against the plan on the account the token belongs to. None of that is built: there is no address, no key and no meter, so nothing can be spent from this page today. Measured The state of the client PENDING The Tingvar client is being built. What this site shows you is the design of it and the measured behaviour of the layer underneath it. There are no product screenshots on this page, because there is no product screen to photograph yet, and a render dressed as a screenshot would be the first thing we asked you not to trust us about. Voice to voice 1471 ms Median, on the bench in us-central1, with a real media server MEASURED 2026-08-19, 122 readings How we measured it Real human sessions 0 sessions Have taken place. Every measured turn so far used pre-rendered audio MEASURED 2026-08-23, 1 reading What happens next The figure on the left is voice to voice: you stop speaking, the first persona's audio starts. It is published with its method rather than as a headline, because a latency number without one is a marketing adjective with digits in it. The figure on the right is why it is not enough on its own. Both sit on [the evidence page](/evidence) with the sample size and the five stages the median breaks into. ## Why from inside an agent ### Handoff and async | | Handoff | Async | | --- | --- | --- | | What the agent sends | A written brief, as the argument to one tool call. | The same written brief. This mode is planned and nothing accepts one yet. | | What comes back | A join URL and the time it expires, then the transcript once the room has ended. | A recorded transcript and no join URL. That is on the roadmap rather than built. | | Whether a human must be present | Yes. A person opens the join URL and speaks in the room. | Nobody would join. That is what is planned, and it is why the mode is on the roadmap rather than offered. | | How long the caller waits | The tool call returns the join URL and ends. The room it opens is one session, 45 minutes. | The call would not return until the debate had ended. It is planned and not built. | | Whether the mode exists today | No. It is the mode being built, and no address has been issued to call it at. | No. It is on the roadmap and no work on it has started. | The two ways of calling this server. Handoff is the mode being built and async is on the roadmap. Scroll to see the rest Everything the async column states is an intention rather than a description. The handoff column is the mode being built. [The roadmap, where async is planned](/roadmap) By the time an agent has written a migration plan or a design document, the proposal already exists in its context, fully formed and long enough that moving it anywhere is a chore. The point at which it is worth putting under pressure is exactly then, before anyone has agreed to it. An agent that has just written a plan has the least incentive to attack it and the least ability to do so. Its context is the argument that produced the plan, so the objections within reach are the ones that argument already answered. Personas that did not write it start somewhere the plan does not reach. The tool call is the point at which the plan stops being agreed with. Up to it, one model has been extending its own reasoning. At it, the same text goes to personas optimising for different things, under no obligation to accept the premise it was written from. What comes back is a transcript rather than a verdict. The disagreement is the output. An agent that wanted a single confident answer would be better served by the model it already has. ### Can I use it from a coding agent or an automation? Tingvar exposes an MCP server, so an agent you already run can convene a room from where your proposal already lives. Claude Code, Cursor, Windsurf and Zed all speak MCP. The connection details, the authentication and an exact list of what exists today against what is still roadmap are on the MCP page, written in the tense each one deserves. A room convened this way is the same room you would join from a browser. ## What a room convened this way has, and what a client has to speak | Capability | Web app | From an agent | | --- | --- | --- | | Convenes a room from a written brief | Yes | Yes | | The same personas, the same arbiter, the same transcript | Yes | Yes | | You speak in the room and interrupt | Yes | Only by joining the room in a browser | | The agent hears the audio | Not applicable | No. It receives the transcript | | Runs unattended, with nobody in the room | No | No. A person opens the join URL and the room starts when they do | | Needs an account | Yes | Yes | What a room convened from an agent has, and what it does not Scroll to see the rest ### What the server does not do - It does not run a debate without a human. open\_room returns a join URL, and the room starts when a person opens it. - It does not stream audio to the client. MCP is request and response, and what a tool call returns is text. - It does not read your files, your repository or your environment. It receives the brief your agent sends it, and nothing else reaches us. - It does not return a verdict or a score, because [disagreement is the output we are designing for](/why-disagreement). [Read about the browser surface](/web) ### The connection contract #### Server address Not issued yet. This page carries no hostname until one resolves, and printing one early would be the first thing on this page a reader would try. #### Authentication An account bearer token, issued from the signed in account settings and scoped to one workspace, and no key issuance is built, so there is nothing to create today. [How the token is issued and scoped](/developers/mcp#authentication) #### Client compatibility No client row is published here, because no client has been tested. A row is earned rather than asserted, and this is what earning one means: - the client named, at the version it was installed at; - that client's own configuration file path, or the command that added the server, exactly as that client documents it; - the operating system it ran on, written down beside the version; - the result and the ISO date of the test recorded against the row; - the tester who ran that client against a live endpoint, named. No row can be added until an endpoint exists to connect to. There is nothing Tingvar specific in the connection, so a client that can already reach a hosted MCP server is likely to reach this one, and likely is not a row. #### Tools INTENDED SHAPE. The server is not live. 2026-09-09 These names are a proposal and they are not ratified. The list the server advertises is the list published here, in full and with each tool signature, in the same commit that makes the server reachable, and it is versioned from that day. | Tool | What a caller gets | | --- | --- | | open\_room | Submits a brief. Returns a join URL, the time it expires, the persona roster and what the session would consume. | | list\_personas | The persona roster: each role, what it optimises for, and the blind spot published against it. | | get\_session | The state of a session you opened: not started, running, or ended. | | get\_transcript | A finished session transcript, with attribution on every line. | | list\_sessions | The sessions opened on your account, most recent first. | | estimate\_cost | What one session would consume against your plan, before anything is spent. | The tools this server is proposed to advertise, and what a caller gets from each Scroll to see the rest #### One authenticated call INTENDED SHAPE. The server is not live. 2026-09-09 ``` # One authenticated call. The address and the token are both placeholders, # because neither has been issued. curl -sS -X POST https:///mcp \ -H "Authorization: Bearer " \ -H "Content-Type: application/json" \ -H "Accept: application/json, text/event-stream" \ -d '{ "method": "tools/list" }' ``` Scroll to see the rest Copy A tool call consumes nothing on its own. It returns a join URL, and one session is consumed at the moment a person opens that URL and the room starts, against the plan on the account the token belongs to. Tingvar plans stop at their included allowance rather than charging past it. Both plans are capped rather than metered, so the monthly figure on the pricing page is the largest amount you can be charged in a month. There is no overage and no silent upgrade. A plan that quietly bills past its own ceiling is a trick rather than a business model. Your transcripts stay available and exportable. ### What you need | Requirement | What it means | | --- | --- | | Transport | Streamable HTTP. It is the transport a hosted MCP server uses, and a client that can already reach a hosted MCP server can reach this one. | | Rate limits | A room is a 45 minute live audio session with a real cost behind it, so the limit is a concurrency limit on rooms rather than a request rate. The number is set when the per session cost is known. | | An account | A room convened from an agent is a room on an account, and the token is issued to that account rather than to the agent. No token can be issued today. | What has to be true before a client can reach this server Scroll to see the rest If a Tingvar session drops for a reason on our side you are not charged for it, and the transcript up to that point is kept and exportable. A dropped room shows in your billing history as a session that was not charged, so you can check rather than take our word for it. Nothing about a dropped room is silently absorbed into your allowance. [What each plan includes](/pricing) ## Questions to answer before you install Where does the brief my agent sends go? To us, and then into every persona system instruction for that session. It is stored with the session, on the account the token belongs to, and it reaches our model vendor as part of the personas' context. [What happens to your data](/trust/your-data) Is my agent context stored? Only what the agent sends as the brief. The server has no access to your repository, your files or your environment, and it asks for none. If your agent sends a whole file, the whole file is the brief and it is stored like any other brief. Who can read the transcript that comes back? You, signed in to the account the token belongs to. Our model vendor receives it as part of the personas' context, and every vendor that touches customer data is named rather than described. [Who at Tingvar can read a transcript, and when](/trust/your-data) What does the server not do? The list is one section up, in full: it runs no debate without a person in the room, it streams no audio to a client, it reads none of your files, your repository or your environment, and it returns no verdict or score. Has a human read what my agent sends? Not necessarily. The brief comes from the agent's context, which can hold text pulled from a file, a page or a tool result that nobody reviewed. What reaches the room is what the agent sends. How do I delete a session convened this way? Yourself, from your account, the same way as a session started in the browser. There is no separate route for one an agent convened and no request to send. [Delete a session or an account](/account/delete) ### Where the recording goes Tingvar does not store your session audio: it is processed while the room is live and no recording of it is kept. What persists is the transcript, the text of what you said and what each persona said, attributed line by line. Only the account that ran the session can read that transcript, and nobody at Tingvar reads one without your written request on a support case. You can delete any session yourself, and deletions are purged from backups within 30 days. The retention schedule is published at tingvar.com/trust/your-data. The brief an agent submits is customer content under the same terms as a brief typed in the browser. [The retention schedule in full](/trust/your-data) [Every vendor that receives a brief or a transcript](/subprocessors) ### Where a subscription is cancelled A room convened from an agent is a room, and it is billed as one on the account it was convened from. There is no store between you and us on this surface: you buy on the web and you cancel on the web from your account settings. The [refund terms](/refunds) carry the window and what happens when a room drops part way through. ## Where to go from here IDLE ## Nobody has used this yet Zero real human sessions have taken place. Every measured turn to date used pre-rendered audio clips rather than a live human. What is built and verified is the decision layer: the turn-taking policy, the bid scoring and the configuration guards, with 619 automated tests passing and 81.78 % combined coverage. Everything there says the system works. Nothing there says anyone wants it, and being first is a real risk you are taking. Start a session See where else it runs --- Source: https://tingvar.com/about Updated: 2026-09-22 # Who operates Tingvar We read the about, legal and footer pages of 8 competing products on 23 August 2026. The number of them publishing a registered entity, a company number and an address together was 0. Ours are below, and every one that is still outstanding is marked as outstanding rather than left off the page. Registered name SNOWWANE LIMITED Company number 13823125 Registered office Online company registered in England and Wales Jurisdiction England and Wales Founder Omotayo Aina Reaching a person [The contact page reaches a person, and states how long a reply takes](/contact) A placeholder above is a fact we do not have yet, marked as one. Each is a row in one register, so a value that arrives replaces every placeholder of that fact on the site in a single change, and a production build refuses to run while any row is still outstanding. Nothing here is a guess dressed as a detail. ## What Tingvar is Tingvar is a live audio room where AI personas argue about your plan. You write a brief first: the proposal you want put under pressure. Three of those personas read it and join a room with you. They do not share a view of what matters most. One optimises for cash survival, one for whether the system holds under load, one for distribution and switching costs. Each carries a deliberate blind spot the others can attack, which is why they argue with each other and not only with you. The assumption two of them price differently is the one carrying your plan's weight. You speak, and whoever is talking stops. Exactly one persona holds the floor at a time, and the room decides who, not the personas. Measured voice-to-voice latency is p50 1,471 ms, p95 1,986 ms, from 122 readings taken on 2026-08-19. Zero real human sessions have taken place. Everything measured so far is a measurement of the machine and not of a conversation. That sentence stays on this page while it is true, and the page that carries the arithmetic behind it is the [evidence page](/evidence). ## The question this page exists to answer ### Who is behind this company? Tingvar is built by a registered company founded by Omotayo Aina, and the entity name, company number and registered address are on the about page. The same identity block sits in the footer of every page, so you never have to open a policy document to find out who you are dealing with. There is a named human, a contact route and a stated response time. ## What we ship, and when we say we shipped it Every change reaches the build log with the date it reached users, good and bad in the same feed. The newest entry is dated 4 September 2026. If that date is ever more than ten days old, it is itself the alarm and you should read it as one. [Read the build log](/changelog) ## Press The name is Tingvar, written in sentence case, with no full stop and no second capital. It is not a suite, a platform or an assistant. Where a category noun is needed, the phrase is a live audio room where AI personas argue about your plan. There is no archive to download. An archive is a file that goes stale the week after it is published and a reader who wants the wordmark wants one file, not seventeen. Write to the contact page and say what you need, and a person sends it. Two things are worth stating before an article is written. There are no customers, no testimonials and no case studies, and the reason is that no real human session has taken place. Every latency figure we publish is a measured one, published with its method on the evidence page, and no other figure has ever been true. [Contact support](/contact) --- Source: https://tingvar.com/faq Updated: 2026-09-12 24 AUG 2026 # Questions and answers Fifty questions, each with one answer. The answers that do not help us are written at the same length and in the same voice as the answers that do, because an honest answer set stops working the moment one of them is shrunk. ## What it is ### What is Tingvar? Tingvar is a live audio room where two to four AI personas argue with you, and with each other, about what is wrong with your plan. You write the plan first. The personas read it before the room opens, and they differ in what they optimise for rather than in what they know. Each one carries a deliberate blind spot the others can attack. It runs in a browser, and your voice stops whoever is speaking. ### Does this just agree with me the way ChatGPT does? No, Tingvar is built to attack the framing of your plan rather than to be helpful inside it, which is the failure a single assistant has by design. Post-training alignment optimises assistants for helpfulness and user satisfaction, so a model validates your framing and then solves the problem inside it. Three personas with opposed objective functions disagree with each other, and that disagreement is the signal. Our target for agreement-only turns is under 5 % of agent turns. It is a target, not a measurement. ### Is this an AI meeting assistant? Tingvar is not an AI chatbot, an AI assistant, a meeting assistant, a note-taker, a summariser, a copilot or an autonomous agent platform. Its output is an argument rather than a record of one. A meeting assistant sits in a meeting you were already having and writes down what happened. Tingvar is the meeting, and nobody in it is on your side by default. If what you need is a transcript of a real conversation with real colleagues, this is the wrong product. ### Why would I want an AI to argue with me? Tingvar's personas attack the proposal and never the proposer, and the useful signal is their disagreement with each other rather than their criticism of you. If a finance persona and a growth persona reach opposite conclusions from the same plan, the assumption they price differently is the one carrying the plan's weight. Every session ends with one question you answer yourself: name one thing you now believe about this proposal that you did not believe an hour ago. An empty answer is a failed session. ### Who is this for? Tingvar is built for someone about to commit to a decision that is expensive to reverse, such as a founder before a raise or a board meeting. The three launch use cases are a pitch rehearsal, a board or exec rehearsal, and a written plan you want put under opposed priorities. It is a poor fit for casual brainstorming, because the personas arrive hostile to a specific document and there is nothing for them to attack without one. ### What are the personas, and are they just different prompts? Tingvar has four personas, Victor Vance, Dr Elena Rostova, Marcus Thorne and Priya Raman, and they differ in what they optimise for rather than in what they know. Victor optimises for cash survival, Elena for whether the system holds under load, Marcus for distribution and switching costs, and Priya for whether the team can ship it. Each one is given a deliberate blind spot the others can attack, which is what stops any single persona reading as the oracle. All four are published on the personas page. ### Why voice, and wouldn't text be easier to read? Tingvar uses voice because interruption is the primary steering control, and interruption is native only in speech. In text you wait for a paragraph to finish. In a room you cut in the moment a persona goes wrong, and it stops. Spoken debate is also faster than text and forces brevity on the personas. You still get the record: a live transcript labels every line with the name of whoever said it, and it is exportable. ### What do I actually get at the end? A Tingvar session ends with a full transcript, every line attributed to the persona who said it, and one question you have to answer yourself. The question is: name one thing you now believe about this proposal that you did not believe an hour ago. The transcript is exportable on every tier. A written post-session brief is a v1.5 item and does not ship today. There is no verdict and no decision memo, because the argument is the output. ## How it works ### Can I interrupt it? Yes, speaking is how you steer a Tingvar session, and your voice stops whichever persona is talking rather than queueing behind it. Human speech preempts everything in the room, and that is an invariant of the design rather than a setting you switch on. Speaking is also how you change the subject, ask a persona to defend a claim, or stop a line of argument that is going nowhere. The client leg of barge-in is not yet measured, and we publish measurements rather than targets. ### Will they talk over each other? Exactly one persona holds the floor in a Tingvar room at any moment, and the room decides who holds it rather than each persona deciding for itself. Server-side voice activity detection is deliberately switched off, so no persona ever decides on its own to speak. Without that, three personas answer every human utterance at once and the room is noise. The floor is held for a bounded period and released, and it can be taken from whoever has it. ### How fast does it respond? Tingvar's measured voice-to-voice latency is p50 1471 ms and p95 1986 ms, from 122 readings over 200 utterances taken on 19 August 2026. p99 is 2156 ms. Voice-to-voice means from you finishing a sentence to the first persona's audio starting. The readings were taken against Vertex AI in us-central1 with a real media server. More than half of that time is the model generating speech, and the stage by stage breakdown is on the evidence page. It is not instant. ### How will I know who is speaking? Each Tingvar persona has a distinct synthesised voice and a fixed position in the stereo field, and a live transcript labels every line with who said it. Separating voices across the stereo field makes rapid speaker changes easier to follow and easier to attribute afterwards. A mono fallback exists, because a single earbud or a laptop speaker makes panning useless. Four personas is the ceiling for the same reason: past four, a listener stops being able to attribute what was just said. ### How many AI personas are in the room? Three personas is the default in a Tingvar session and four is the ceiling, because past four a listener stops being able to attribute what was just said. The default three are Victor Vance, Dr Elena Rostova and Marcus Thorne. Priya Raman is the optional fourth, and she is the one to add when the question is whether the team can actually ship the thing rather than whether the plan is right. There is no setting that puts eight voices in a room. ### Do I have to prepare anything? Yes, Tingvar asks for a written brief before the room opens, and sessions that start from a written proposal are markedly better than cold starts. The brief is the proposal you want put under pressure: what you are proposing, what it costs, what you are assuming and what would make it fail. It is a document the personas have already read when they arrive. The cold start comparison is our own observation from internal testing, not a measurement. ### How long is a session? A Tingvar session is 45 minutes by default, and that is the unit everything on the pricing page is quoted in. You can leave a room early. Nothing keeps running after you go, and a session you end early still produces the transcript up to that point. Ending your first session inside the first five minutes is also how you claim the Pitch-Ready guarantee. ### Do I need headphones, and where am I supposed to do this? Tingvar needs headphones and a room where you can talk out loud for 45 minutes, because it is a spoken conversation and you will be interrupting people. Open plan is the honest problem here, and we do not have a good answer for it beyond a closed room or a quiet hour. A mono fallback exists for a single earbud, though you lose the stereo separation that makes speakers easier to tell apart. This is a hypothesis rather than something we have watched people do. ## Cost and billing ### What does one session cost me? Tingvar is priced by the plan rather than by the session, and a session is 45 minutes with three personas. Professional is 149 USD a month for ten sessions and Team is 399 USD a month for thirty, both capped, so the monthly figure is the largest amount you can be charged in a month. There is no overage and no credit system to convert into money. An Enterprise tier is priced with you. ### Is there a free tier? Tingvar does not have a free tier, and there are two paid plans instead. Professional is 149 USD a month and Team is 399 USD a month, with an Enterprise tier priced with you. What we offer in place of a free tier is the Pitch-Ready guarantee: start your first session, and if five minutes in you do not feel cross-examined, end it inside that window and we refund the payment in full. ### What happens when my plan's sessions run out? Tingvar plans stop at their included allowance rather than charging past it. Both plans are capped rather than metered, so the monthly figure on the pricing page is the largest amount you can be charged in a month. There is no overage and no silent upgrade. A plan that quietly bills past its own ceiling is a trick rather than a business model. Your transcripts stay available and exportable. ### What happens if the room drops halfway through? If a Tingvar session drops for a reason on our side you are not charged for it, and the transcript up to that point is kept and exportable. A dropped room shows in your billing history as a session that was not charged, so you can check rather than take our word for it. Nothing about a dropped room is silently absorbed into your allowance. ### Can I get surprised by an overage charge? No, a Tingvar plan either stops at its included allowance or charges a published per-session rate you have already seen, and there is no third behaviour. Which of the two applies to your plan is stated in the same viewport as the price, not in a footnote and not in the terms. The reason we write this down is that metered pricing in this category produces the same complaint every time, which is a bill nobody could have predicted from the pricing page. ### How do I cancel? You cancel Tingvar yourself on the web in three clicks or fewer from your account settings, with no retention interstitial and nobody to talk to first. The control is at the account cancel page and it is linked from account settings. Your plan runs to the end of the period you have paid for, and your transcripts stay exportable until you delete them. Deleting your account cancels the subscription at the same time, and the account page says so before you confirm rather than after. ### Can I get a refund? Tingvar refunds your first payment in full if you end your first session inside the first five minutes. That is the Pitch-Ready guarantee, and you trigger it from your account without explaining yourself. Carry the session past five minutes and it is not refundable, because by then you have the argument, the transcript and the export. Your statutory withdrawal right in the UK and the EU is separate, is not limited to five minutes, and is not ours to remove. ## Privacy and your data ### What happens to my recording? Tingvar does not store your session audio: it is processed while the room is live and no recording of it is kept. What persists is the transcript, the text of what you said and what each persona said, attributed line by line. Only the account that ran the session can read that transcript, and nobody at Tingvar reads one without your written request on a support case. You can delete any session yourself, and deletions are purged from backups within 30 days. The retention schedule is published at tingvar.com/trust/your-data. ### Who can hear my strategic weaknesses? Nobody else joins a Tingvar room: the participants are you and the AI personas, and the audio is not stored for anyone to play back later. The transcript is the part that persists, and it reaches our model vendor as part of the personas' context. Every vendor that touches customer data is named on the sub-processors page, including the model vendor. Who at Tingvar can read a transcript, and in what circumstances, is stated at tingvar.com/trust/your-data. ### Is my data used to train AI models? No, Tingvar does not use your brief, your transcript or your audio to train models. Google, which runs the model, states that customer data submitted to Vertex AI is not used to train or fine-tune its models without permission, that cached content is held for up to 24 hours in the data centre serving the request, and that prompts may be logged for abuse detection unless an exception is agreed. It is linked from our sub-processors page so you can check it. ### Who are your sub-processors? Tingvar publishes every vendor that touches customer data at tingvar.com/subprocessors, including the model vendor, with the jurisdiction each one operates in. The list names the underlying cloud for any vendor that is itself reselling infrastructure, because a list that stops at the reseller tells a reviewer nothing. We give 30 days' notice before a new sub-processor begins processing customer data, and you can subscribe to changes. The page carries an effective date, so you can tell how current it is. ### Where is my data stored? Your Tingvar session audio and transcripts are processed in the United States, in Google Cloud's us-central1 region, and we do not currently offer an EU-only processing region. For customers in the UK and the EEA, that means a transfer out of your region. We rely on Standard Contractual Clauses and the safeguards in our DPA. If your procurement requires EU data residency, the honest answer today is that we cannot meet it, and we would rather you knew that before you bought than after. ### How do I delete a session? You delete a Tingvar session yourself from the session list, and deleting your account destroys every session in it. Deletion is self-service and does not go through support. Anything you delete is purged from backups within 30 days, and we publish that number because an unbounded backup window is what fails a security questionnaire. Billing records survive deletion for seven years, because tax law requires it, and we say so rather than implying everything vanishes. ### I am going to describe things about my company that are not public. Should I? A Tingvar session is a spoken record of what is wrong with your plan, which makes it one of the most sensitive artefacts your company could produce. So the honest advice is to read our data page before your first session rather than after it. What we control is where it goes, who can reach it and how fast you can destroy it, and all three are published. Names, unreleased figures and anything under an agreement with somebody else are worth redacting from the brief. ### Will it invent numbers about my business? A Tingvar persona once turned the misheard phrase "turns in" into a confident objection about 15 % monthly churn that had never been said. We publish this because a room whose value is pressure-testing your numbers, and which invents numbers, is worse than no room at all. The personas argue from the brief you write and from what you say, and they cannot see your accounts or anything you have not told them. If a number sounds wrong, say so out loud. ## Platforms and requirements ### Is there a web version? Yes, Tingvar runs in a browser with nothing to install, and that is the primary way to use it. You need a current browser with WebRTC and Web Audio support, a working microphone, and a connection that can hold a voice call. The browser asks for microphone permission when you join a room, and not before. Everything the product does in v1 is available in the browser, so nothing is held back for an app you have to download. ### Is there an iPhone or Android app? Tingvar does not have an iOS or Android app on the stores today, and the browser version runs on both platforms in the meantime. Both apps are planned and both are P0 on our own list. We do not publish a store badge for a listing that does not exist, and we do not publish a design render dressed as a screenshot. The iOS and Android pages state what is true today, and you can ask to be told when either one ships. ### What do I need to run it? Tingvar needs a current browser with WebRTC and Web Audio support, a working microphone, and a connection that can hold a voice call. Chrome, Edge, Safari and Firefox on their current versions all carry what it needs. Headphones are worth having, because a laptop speaker feeding back into a laptop microphone is a reliable way to make any voice product unpleasant. A poor connection costs you audio quality first. ### Can I use it from a coding agent or an automation? Tingvar exposes an MCP server, so an agent you already run can convene a room from where your proposal already lives. Claude Code, Cursor, Windsurf and Zed all speak MCP. The connection details, the authentication and an exact list of what exists today against what is still roadmap are on the MCP page, written in the tense each one deserves. A room convened this way is the same room you would join from a browser. ### Can I use it with a screen reader, or without hearing the audio? Tingvar targets WCAG 2.2 Level AA on the website, and the live transcript in the room is the media alternative for anyone who cannot use the audio. Known gaps are listed by name on the accessibility page with the date each one was found, rather than summarised as a commitment to inclusion. The product itself is a spoken argument, and we will not pretend that a transcript makes it equivalent for a deaf user. There is a route to report a barrier and a stated response time. ## How it compares ### How is Tingvar different from ChatGPT? Tingvar and ChatGPT solve different problems: ChatGPT gives you one assistant optimised to be helpful, and Tingvar gives you three personas optimised to disagree with each other. You can interrupt a Tingvar room out loud and the personas can interrupt each other, which a chat window does not do because there is nothing to interrupt. Most people who use Tingvar will keep using ChatGPT, and for most of what they do that is the right choice. The comparison page states where each one wins. ### How is this different from three ChatGPT tabs? Three ChatGPT tabs give you three answers, and a Tingvar room gives you an argument, because separate tabs cannot hear each other. Nothing in tab two knows what tab one said unless you paste it there yourself, and by then you have chosen which criticism survives. Inside one chat window the instruction to be critical decays after two or three turns. In a room, one persona attacks another persona's reasoning while you listen, and you can cut in the moment either of them goes wrong. ### How is Tingvar different from AI Group Call? AI Group Call runs on iPhone and Android only, and Tingvar runs in a browser with nothing to install. Both are live voice products with more than one AI participant, which makes this the closest comparison in the category. AI Group Call meters by the minute, and Tingvar prices by the 45-minute session. Tingvar publishes a measured voice-to-voice latency figure with its method and its date. The full comparison, with the crawl date on every fact, is on the comparison page. ### How is Tingvar different from AskVerdict? AskVerdict's product is text and Tingvar's is speech, so their personas cannot interrupt each other and you cannot interrupt them, because there is nothing to interrupt. Where they are ahead of us, stated here rather than left for you to find: AskVerdict ships an API, an SDK and a command line tool, and publishes seven case studies. We have none of those. What we have is a spoken room you can cut into and a measured latency figure. ### Could we build this ourselves with CrewAI or LangGraph? You could build multi-agent text debate in an afternoon, and the hard part of Tingvar is none of that. The hard part is deciding which of three simultaneously willing speakers gets the floor, fast enough that the room still sounds like a conversation, without any of them talking over the human. Agent level turn-taking, where each agent decides for itself whether to speak, stops working at three agents. Our arbiter is a room level policy, and it is where the latency budget is spent. ### What is Tingvar bad at? Tingvar is bad at anything you want agreement or output from: it will not draft your deck, write your code, or summarise your meeting. It is a poor fit for casual brainstorming, for a decision you have already made, and for anyone who wants their instinct confirmed. It has no published customer outcomes, because zero real human sessions have taken place. The evidence page carries what we have measured and what we have not, including the rows that are empty. ## Enterprise and security ### Do you have SOC 2 or ISO 27001? No, Tingvar holds no SOC 2 report and no ISO 27001 certificate, and we will not claim a stage of either one that we have not reached. No product in the set we crawled on 23 August 2026 publishes one either. What we do have is published: the controls on the security page, a pre-signed DPA you can read in full without a form, and every sub-processor named including the model vendor. We answer a security questionnaire in five business days. ### Do you have a DPA? Yes, Tingvar publishes a pre-signed data processing agreement at tingvar.com/dpa that your legal team can read in full before speaking to anybody here. It carries the sub-processor list by reference, the 30 day change notice, and the transfer mechanism we rely on for customers outside the United States. There is no form in front of it and no call required. A signed file follows once the four outstanding entity details are supplied. If your own paper has to be used, the demo page is the route. ### Can we get single sign-on? No, Tingvar does not have single sign-on today, and the enterprise scaffolding it belongs to has not been built. Single sign-on, tenancy controls and audit export are on the roadmap in the future tense, each with the state it is in, the date it was last reviewed, and no delivery date. We would rather you learned that here than on a call three weeks into a procurement. If it is a hard requirement, tell us on the demo form and we will say honestly where it sits. ### What is your uptime commitment? Tingvar does not publish an uptime percentage, because we have no operating history to base one on. A number invented before the service has run is not a commitment, it is a guess with a percentage sign on it. There is no status page yet: operational notices and incident updates go out through support, and a status page is scheduled for general availability rather than for launch. When we have enough operating history to write a service level agreement we can keep, we will publish it and date it. ### Who is behind this company? Tingvar is built by a registered company founded by Omotayo Aina, and the entity name, company number and registered address are on the about page. The same identity block sits in the footer of every page, so you never have to open a policy document to find out who you are dealing with. There is a named human, a contact route and a stated response time. ## Where we actually are ### Do you have any customers? Tingvar has no customers, no testimonials and no case studies, because zero real human sessions have taken place. Every measured turn so far used pre-rendered audio, which means we have measured the machine and not the conversation. There is no logo wall on this site, no counter and no invented outcome figure, and there will not be one until there is something real to put in it. We will publish the first real session in full, with consent, when it happens. ### Has anyone actually used this? Zero real human sessions have taken place. Every measured turn to date used pre-rendered audio clips rather than a live human. What is built and verified is the decision layer: the turn-taking policy, the bid scoring and the configuration guards, with 619 automated tests passing and 81.78 % combined coverage. Everything there says the system works. Nothing there says anyone wants it, and being first is a real risk you are taking. ### What have you actually measured? Tingvar has measured voice-to-voice latency and the automated test suite, and has measured nothing at all about whether a session changes anyone's mind. The measured figures are p50 1471 ms and p95 1986 ms voice-to-voice, from 122 readings over 200 utterances on 19 August 2026, and 619 automated tests passing at 81.78 % combined coverage. The three numbers that would tell you whether it does its job have targets and no measured values, and those rows are left visibly empty on the evidence page rather than removed. ## Something here is wrong or missing Every answer above is one node in a register, and the same node is what the machine readable copy of this page carries. If an answer is wrong, write to [corrections@tingvar.com](mailto:corrections@tingvar.com). We correct it within 1 business day and log the correction publicly. Start a session