Free while it's
a beta.
There is no card to enter, and no billing in the app to enter it into. There is no API key to enter either: every model runs on our keys, inside 200 credits a day per licence. What Troctor costs after the beta depends on a feature we have not built yet, so this page gives you the reasoning instead of a number we would have to walk back.
Private beta
Free, invite only, while we work out whether this is genuinely useful.
- Unlimited meetings, imports and searches
- 200 credits a day per licence, on our keys
- Three models, no API key and no vendor account
- One licence key, three machines
- Every feature that exists today, no card
What a day of it buys
Credits rather than questions, because the models behind them do not cost us the same.
- Groq Turbo: 1 credit an answer
- GPT (fast): 2 credits an answer
- Claude Haiku: 4 credits an answer
- Grok 4.6: 6 credits an answer
- GPT (deep): 8 credits an answer
- So 200 answers a day, or 25 of the dearest
- Counted per licence, resets at midnight UTC
Live mode, when it is proven
Transcribing a meeting is a bill somebody has to pay, and it is now ours rather than yours, which is why it has a ceiling.
- Everything above
- No daily limit on listening during the beta
- Our vendor account, our key, $0.12 per channel hour
- Unproven, on the roadmap
The middle card is what the beta gives you, not a plan you can buy, and the figure on the third is a vendor's list rate we pay rather than a price to you. Today the app has no payment screen, no subscription and no meter running in the background. It has a licence key, and during the beta the key is free.
Why there isn't
a number yet
The usual way to price software is to look at what comparable software charges and land somewhere nearby. We would rather start from what Troctor actually costs to run, and right now that figure is close to nothing, because almost every expensive part of it happens on hardware you already own.
Search is free because it runs on your Mac
Importing a transcript, writing it out as Markdown, chunking it by speaker turn, embedding it and searching the result all happen locally. The embedding model is a small open one that downloads once, about 33 MB, and then stays put. The index is a SQLite file sitting next to your vault. The insights, contradictions between meetings, commitments nobody followed up, themes that keep returning, come out of SQL queries over that index rather than out of a model.
So none of that touches a server we rent. You can import four years of transcripts and run two hundred searches in an afternoon and our costs do not move. The parts that do move are the answers and, when it is switched on, the transcription, which is why each has a daily allowance on it and why the allowance is a number we can change: those are the only two lines in this product with a bill attached, and we would rather bound them honestly than pretend they are free to us.
Every answer is on our key, inside a daily allowance
There is no longer a model that runs on a key of yours, because the app has nowhere to keep one. It sends a model id to our backend; the backend holds every vendor key and decides which upstream answers. What bounds that is 200 credits per licence per day, resetting at midnight UTC and counted per licence rather than per machine, so three Macs on one key share it. When it runs out, Troctor says so and tells you when it comes back, and nothing is sent anywhere in the meantime. There is no way to raise it, because there is no key to add.
Credits rather than a count of answers, because the models stopped costing the same. Groq Turbo costs one credit an answer, GPT (fast) costs two, and GPT (deep) costs eight. So a day is 200 answers on the free one, 100 on GPT (fast), or 25 on the dearest, and the picker states the price beside each model before you choose it. A flat count of answers would have let one person spend eight times what the person beside them spends by picking a different line in a dropdown, with nothing in the product noticing.
That arrangement exists for one reason. Making a stranger open an account with a model vendor, add a payment method and generate a key before Troctor will answer a single question is the largest drop-off there is, and it happens before the product has shown them anything. Paying for the answers is cheaper than losing the people who would have stayed. What is left is: install it, paste the licence key we emailed you, start asking.
It is not free of cost to you, and the cost is not money. Every request goes through our backend, so your question and the retrieved excerpts pass through a machine we own. We relay them and store none of them, and the privacy policy lists exactly what is written down. Until recently you could take us out of that path by pasting a key of your own. You cannot now, and that is the honest trade for never having to find a key at all.
The model list itself comes from our server rather than from the build you installed, so a model can be added, replaced or switched off without you updating anything. Four are offered in the beta and they are the four above.
Which leaves the app itself
That is the honest thing to charge for. Not your questions, not your hours, not somebody else's tokens with a fee stapled on. The app, and the ongoing work of keeping retrieval accurate, the parsers current and the vault format stable. There are sensible numbers for that. We do not yet know which one, and the fastest way to find out is to watch how people use the beta rather than guess in public and defend the guess afterwards.
Live mode is what changes the shape, and here is the number
Everything above is cheap to run because it runs on your Mac. Live mode is not, because transcription happens at somebody else's company and they charge by the length of the connection.
Troctor's live mode is built around Soniox, and the account is now ours rather than yours. The app asks our backend for a key, the backend mints one that lasts sixty seconds and can do nothing but open a stream, and the audio then goes from your Mac straight to Soniox without passing through us. The bill comes to us, which is exactly why there is a ceiling on it. The figures, from their list pricing rather than from a promotion:
- There is no free tier. Soniox removed theirs in October 2025. It is pay as you go, which is why the allowance below is a real limit rather than a formality.
- $0.12 per channel hour. Troctor keeps your microphone and the other people on separate connections, because that is what makes a transcript say who said what without guessing.
- So about 24 cents an hour for both sides of a meeting, on our card rather than yours.
- No daily limit during the beta. There was one, of 240 minutes, and it was removed because it was never the thing that stopped anybody and it showed testers a countdown to a wall. What remains is a cap of one hour on a single meeting, a limit on how many connections a licence may open in a day, and a ceiling on how much transcription we will authorise in a day, which ordinary use does not come near. A meeting capturing both sides uses two connections, because that is what makes the transcript say who said what.
Two things about that rate are worth knowing. It is billed on how long a connection is held open, silence included, not on how much anybody said: a quiet hour costs the same as a busy one, which is why Troctor closes a session on every path it can and stops at a two hour cap. And the app reports the seconds it has spent while the meeting is still running rather than at the end, because the sessions that cost the most are the ones that never reach a clean stop.
We are not charging for live mode and there is nothing here to buy. If that changes, this page will say so before it happens. Live mode itself is on the roadmap as built and unproven, and the security page is where the privacy side of it is set out.
Three things we will not do
Whatever the eventual price is: your vault will never be held behind it, because your vault is plain Markdown in a folder you chose and we have no way to reach into it. Nothing you have already imported or already asked will be billed retroactively. And anyone who was in the beta will hear about a paid plan from us before they hear about it from a locked screen.
What costs what
| Where it runs | Who pays | |
|---|---|---|
| Importing transcripts | Your Mac | Nobody |
| Writing the Markdown vault | Your Mac | Nobody |
| Embedding and indexing | Your Mac, local model | Nobody |
| Search and retrieval | Your Mac, SQLite | Nobody |
| Insights and contradictions | Your Mac, SQL only | Nobody |
| Writing an answer, free models | Our backend, then Groq or Google | Us, 1 credit of 200 a day |
| Writing an answer, GPT (fast) | Our backend, then OpenAI | Us, 2 credits of 200 a day |
| Writing an answer, Claude Haiku | Our backend, then Anthropic | Us, 4 credits of 200 a day |
| Writing an answer, Grok 4.6 | Our backend, then xAI | Us, 6 credits of 200 a day |
| Writing an answer, GPT (deep) | Our backend, then OpenAI | Us, 8 credits of 200 a day |
| The app and its updates | Your Mac | Free during the beta |
| Live transcription of a meeting | Your Mac straight to Soniox | Us, with no daily limit during the beta |
| Recording a meeting to disk | Your Mac, about 115 MB per channel hour | Nobody, and nothing is uploaded |
The beta offers Groq Turbo, and GPT (fast) and GPT (deep) on OpenAI. The app shows what each one costs against the day's allowance before you pick it, and how much of today is left. The list comes from our server, so it can change without you updating anything.
The ones about money
What does the beta cost?
Nothing, and there is nowhere to pay even if you wanted to. No card form, no subscription, no trial countdown. It is invite only: join the waitlist on the download page, we send you a licence key, you enter it on first launch and it works on three machines.
What will it cost later?
We do not know, and we are not going to make one up so this page looks finished. What we can tell you is the shape. The thing being sold will be the app rather than your usage, because most of your usage runs on your own hardware and does not cost us anything. The two exceptions are answers and transcription, and both of those are on our vendor accounts now, which is why both carry a daily allowance rather than a bill.
What does live transcription cost?
It costs us about 24 cents an hour for a meeting with both sides captured. Soniox list at $0.12 per channel hour and Troctor uses one connection per channel, because that is what makes the transcript say who said what. Soniox has no free tier, and has not since October 2025, so this is a real bill rather than a rounding error, which is why a licence has a ceiling on how much transcription we will authorise in a day and on how many connections it may open. The daily minute allowance that used to sit here was removed during the beta: it was reported by the app rather than enforced, so it bounded nothing, and all it did was show testers a number counting down. It costs you nothing and there is no account for you to open. Billing is on connection time rather than on speech, so silence costs the same as talking, which is why sessions are closed on every path and capped at an hour.
Is there a free tier with a model included?
Yes, and there is nothing else. Every model runs on our keys, and a licence gets 200 credits a day, which is 200 answers on either of the free models or 25 on the dearest one. There is no API key to find first, and no way to add one. The trade is that every request goes through our backend to reach the vendor, so your question and the retrieved excerpts pass through infrastructure we run. We store none of it, and the privacy policy is specific about what we do write down.
Who pays the model bill?
We do, all of it, on our own accounts with Groq, Google and OpenAI. The app cannot pay it even if you wanted it to: it holds no vendor key, and the code that used to store one was deleted. What stops that being unbounded is the daily allowance rather than a card of yours.
What happens when I run out of credits?
Troctor tells you, says when the allowance returns, and stops. If a cheaper model can still afford the question, it says that too. Nothing is sent anywhere, and nothing is queued up to be sent later. Importing, indexing, searching and reading old answers all keep working, because none of those need a model. There is nothing you can add to lift the limit, and it resets at midnight UTC.
Does asking more questions cost more?
Searching does not. Only the final answer touches a model, and it receives your question plus whatever you attached to that meeting and what has been said in it, not your whole vault. What it costs is credits rather than money: one on either free model, two on GPT (fast), eight on GPT (deep), out of 200 a day.
What happens if I stop using Troctor?
Your vault carries on existing. It is a folder of Markdown files with YAML frontmatter, in a location you picked, readable in Obsidian, VS Code or TextEdit. Delete the app and every meeting note is still sitting there. The SQLite index is derived from those files, so even that can be rebuilt from what you kept.
Do I lose my data if I do not take a paid plan?
No, and the architecture is what makes that a fact rather than a promise. Your vault is never uploaded and we store nothing from your meetings, so there is nothing on our side to switch off. The worst a future paid plan could do is stop the app from opening, and stop the free answers. It could not reach into a folder on your disk, and we would not want to be the kind of company that could.
Do you have team pricing?
Not yet. There is no shared billing because there is no billing, and a team version raises questions about shared vaults that we have not answered properly. If you are thinking about Troctor for a team, write to us and tell us what you need. That is more useful to us right now than a signup.
Free is easy to evaluate.
Import an afternoon of old transcripts and ask it something you would have had to go digging for. If it is not useful, you have lost twenty minutes and no money.
Apple Silicon · macOS 12 Monterey or later · not notarised yet, so first launch takes one extra click