Pricing

Hosted by us, or on your own infrastructure.

The same binary either way, and every feature in every plan. Hosted: free to start, then a flat monthly fee by plan size plus prepaid AI credit — never per seat, bought in the console in about a minute. Deployed: the open-source binary yourself for nothing, or installed by us on your premises.

Hosted by us

Free to start. Then a flat fee by plan size, plus AI credit.

Everybody in the channel can use it, so billing a seat at a time would be charging you for the thing that makes it work. Pro is a flat monthly fee set by the plan size — a band, not a seat count, and nobody is turned away for going over it — with prepaid AI credit for model spend on top: pick the size in the console, pay by card, and it is on. Enterprise is the same shape sized for a company — and that one is a conversation, because the paperwork is the part that needs people.

Free

The hosted service, with $5 of model spend a month on the included endpoint and everything on the list below. No card. It does not expire.

$5 a month of model spend, included
Pauses at the ceiling, never bills past it
Every call logged with input tokens, output tokens and cost, so the spend is yours to control
Everything in the list below

Try it now

Pro

From $49

A flat monthly fee by plan size — how many people use the bot in a month, counted rather than declared — plus prepaid AI credit for model spend, metered at the provider’s price with no margin. Pick a size, pay by card, top the credit up whenever you need to. Nobody is refused for going over: the console shows the count and you move up when it suits you.

5 sizes, priced below — counted in the people who use the bot, never sold per seat
Prepaid AI credit at the provider's price, topped up in the console any time
A European region on request, if your data may not leave the EU
An engineer sets it up with your team, then reviews the usage with you
Everything in the list below

Already have an account? Settings → Billing, pick your size, pay by card. About a minute, and nobody has to call you.

Start free, then choose a size

Enterprise

Quoted

Everything in Pro, sized for a company — and the paperwork a company needs before it can say yes.

A flat fee and AI credit sized to the company
One invoice, purchase orders, many workspaces under one account
A signed DPA and the security review answered in writing
A European region on request, if your data may not leave the EU
Bring your own model key instead of AI credit, on request
A named engineer to call: requirements with the team, setup end to end, usage reviews after
Everything in the list below

Talk to us

Pro, by size

A size is how many people may use the bot in a month. We count the ones who actually did, over the last 30 days, and nothing is refused for going over: the console shows the figure and you move up when it suits you. Model spend is separate, at the provider’s price: every fee carries the credit shown beside it, granted again at each renewal. Past 250 people, talk to us.

Up to 10 users: $49 a month$5 AI credit included
Up to 25 users: $99 a month$10 AI credit included
Up to 50 users: $199 a month$20 AI credit included
Up to 100 users: $349 a month$35 AI credit included
Up to 250 users: $599 a month$60 AI credit included
More than 250 users: Talk to us

Billed monthly to a card and cancelled on the same screen. The credit a size includes expires with the month it came with; credit you buy does not. Credit is bought in $100 to $1,000 amounts, or any figure you type. Prices are in US dollars and exclude tax, which is added at payment where it applies.

In every plan

Unlimited channels and threads
Every native tool: Slack history and search, documents, web, artifacts, the JavaScript sandbox
Unlimited connections, bundles, domains and skills
Remote MCP servers, with bearer or OAuth sign-in
Its own MCP server and developer API, for Claude, Codex, Cursor and scripts, with each person's own access
Tool packs for ClickUp, GitHub, GCP, Sentry and HubSpot
Google Workspace on each person's own account: mail, calendar, Drive, contacts
Memory and scheduled routines, per channel
Connected repositories and fix jobs with draft pull requests
The admin console, with roles, invites, two-factor sign-in and CSV export
Approval tiers and access requests
Per-workspace, per-channel and per-user budgets and limits
Full audit trail of turns, tool calls, proxied requests and spend

Model spend

At the provider’s price, against a ceiling.

Every completion’s tokens and cost are recorded at the provider’s own price, and that is the number that counts against your budget — there is no margin on tokens. On the default open-weight model, a typical answered question runs to a fraction of a cent.

A question answered from a thread$0.002

A document-grounded policy answer$0.001

A fix job: clone, tests, change, PR$0.18

Illustrative, on google/gemini-flash via OpenRouter. Your own figures are in the console from the first turn — nothing there is an estimate.

Getting it used

The licence is the easy part. Adoption is where the money goes missing.

Most companies are already paying for AI and getting thin results from it, because a tool arrives installed and nobody’s work changes. The fix is not another tool: it is joining the ones you already pay for to the place your people already talk. So every plan with a person attached to it comes with that person — an engineer, on your workspace, who does this and then checks that it took.

01We talk to the team first

An engineer meets the people who will type at it, not only whoever signed. Which channels they live in, which tools they already pay for, which questions get asked twice a week, and what nobody may do without a human saying yes.

02We wire it up end to end

Slack or Microsoft Teams, Google Workspace, your repositories, ClickUp, Sentry, HubSpot and your own MCP servers, joined into one place: connections, bundles, approval tiers, skills, routines and budgets, configured in your workspace by us. Not a starter template and a link to the documentation.

03We stay until it is used

Every turn, tool call and dollar is already in the console. We go through those numbers with you over the first weeks — what got used, which questions were answered badly, which channel never started — and change the setup until it lands. On your own infrastructure the numbers never leave it; we read them with you.

Included in Pro, in Enterprise, and in an install on your own premises — a deployment we hand the keys to is still one we run the rollout for. On the free plan or the open-source binary, ask for it on its own: talk to us.

Where it runs

Hosted in the EU, if that is the sentence in your policy.

The hosted service runs in the United States by default. If your data may not leave the EU, we run the same binary as your own instance in a European region — Frankfurt, Belgium, or whichever one your policy names — with its database and its documents created in that region. Nothing about the product changes: every feature in the list above is the same one.

On Pro and on Enterprise it is on request, and the reason is the same on both: a European region means an instance of its own rather than a switch we flick, so we price it with you. On a deployment you run yourself the region was always yours — attest_tag runs wherever you install it.

What actually crosses a border

Your database, documents and audit log: The region you picked. They are created there and they do not move.
Your Slack workspace: Slack's, not ours. Channels are read on each turn rather than mirrored, so most of what you would worry about never lands in our database at all.
Your Microsoft Teams organisation: Microsoft's, with one difference: Teams gives an app no way to read a conversation back, so the bot keeps its own record of the conversations it is in — in your database, in the region you picked.
Model calls: Wherever the endpoint you point at lives — which is the part a region alone does not settle. Name an EU endpoint, or your own, and prompts stay inside it.

An EU deployment answering on a US model endpoint still sends the prompt across. Tell us the endpoint and we wire that one.

Why nothing is withheld

Nothing here is held back to sell you a bigger tier.

Approval tiers, audit export, budgets, two-factor sign-in and Slack sign-in are the things a tiered vendor puts behind “Enterprise”. They are also the things that make this safe to run at all, so they are in every plan — the free one and the open-source binary included. What Pro and Enterprise add is scale and paperwork.

Questions

About the bill

Do you charge per user?

No. Pro is a flat fee by plan size — up to 10 users, up to 25 users, up to 50 users, up to 100 users, up to 250 users, more than 250 users — plus prepaid AI credit for model spend, and the free plan has no size at all. A size is the number of people who used the bot in the last 30 days, counted rather than declared, and nobody is refused for going over it: the console shows the count and you pick a larger size when it suits you. Adding a channel or a thread never changes what you pay. The figures are in the table above.

What is AI credit?

Model spend, paid for in advance. Every completion’s tokens are recorded at the provider’s own price and drawn from the credit; the console shows the balance beside the month’s spend, the bot pauses when it runs out and says so in the channel rather than going quiet, and you top it up with a card on the same screen. There is no margin on it, and it does not expire.

Is the free plan a trial?

No — it does not expire. Every organisation starts on it, with $5 of model spend a month on the included endpoint and everything else unlimited. When the month’s spend reaches the ceiling the bot pauses and the console says so, with the size picker one click away.

What is the difference between Pro and Enterprise?

Scale and paperwork, not features. Pro is a flat fee by plan size plus prepaid AI credit, bought in the console with a card. Enterprise is the same shape sized for a company, on one invoice, with a signed DPA, the security review answered in writing, and a named person to call. Every feature is in every plan.

Do you help us actually get it used?

That is the point of the paid plans. An engineer talks to the team first — the people who will type at it, not only whoever signed — and writes down the channels, the tools you already pay for, the questions asked twice a week and the things nobody may do without a human saying yes. Then we configure all of it in your workspace: connections, bundles, approval tiers, skills, routines and budgets. Then we go through the console’s usage with you over the first weeks and keep changing it until it lands. It is not an install and a welcome email. The same applies to an install on your own premises, where the numbers stay on your infrastructure and we read them with you.

Can you host us in the EU?

Yes, on request on Pro and on Enterprise. The hosted service runs in the United States by default; ask and we run your instance in a European region instead, with its database and documents created there. It is an instance of its own rather than a switch, so it is quoted; on a deployment you run yourself you choose the region and there is nothing to ask. The thing to decide alongside it is the model endpoint — completions go wherever the endpoint you point at lives, so an EU instance answering on a US endpoint still sends the prompt across. Name the one you want and we wire that one.

How does an upgrade work?

In the console: Settings → Billing, pick the size that covers the people who use the bot, pay by card. Stripe takes the payment and the workspace is on Pro on the way back — no call, no order form, no waiting on us. Credit is bought on the same screen, and the card lives in Stripe’s own portal rather than with us — which is also where you move to a different size later, or cancel. If you would rather have an invoice, a purchase order or a signed DPA, that is Enterprise, and that one is a conversation.

Do you mark up model tokens?

No. Every completion is recorded at the provider’s own price, and that is what is drawn from a plan’s AI credit. Deployed on your own infrastructure, you point it at your own endpoint and pay your provider directly — we never see that invoice. On Enterprise you can have that arrangement on the hosted service too: bring your own model key on request, and the AI credit line leaves the quote entirely.

What stops a runaway bill?

Two limits, and the console says which one is binding. Prepaid credit is the hard one: the bot cannot spend money you have not already paid, so the worst case is a stopped bot rather than an invoice you did not expect. On top of that you set a monthly budget for the workspace and another per channel, and a per-user rate limit caps requests per hour. When a limit trips, an alert goes to the channel you nominate, once an hour. The failure mode is a stopped bot and a message — not a surprise at the end of the month.

Can we get a refund?

Unused credit comes back. Write to [email protected] and we return it to the card that bought it, less anything already spent — and we do it without being asked if you stop. The monthly fee is not refunded part-way through a month: cancel in the console and Pro runs to the end of the period you have paid for, then stops renewing. Nothing is charged after that.

What if our card fails?

Stripe retries it and emails your billing address. Pro stays on while they retry; if it is still failing after that, the workspace goes back to the free plan and keeps whatever credit it had already bought. Nothing is deleted, and the bot keeps answering inside the free plan’s ceiling.

What does “we deploy it on your premises” include?

The install, on infrastructure you own: your cloud account, your cluster or your own machines, with your own Slack or Teams app, model endpoint and keys. We run the first update with you and hand over the runbook. Afterwards it is yours, under the same MIT licence, and rerunning the platform script is how you ship a new version — the self-hosted page is the same runbook, for doing it yourself.

What happens to our data if we stop?

Remove the app from Slack and the bot is gone from your workspace, which was always read on each turn rather than mirrored. In Microsoft Teams, where the bot keeps its own record of the conversations it was in, that record is part of the organisation’s data. On the hosted service, write to [email protected] and the organisation’s data is deleted. On your own infrastructure, it is your database and your documents folder.

Start on the free plan, or clone the repository.

No card, no call, and the ceiling is a number in the console rather than a surprise at the end of the month.

Try it now Deploy it yourself

Pro is in the console — pick a size, pay by card. Enterprise, or an install on your premises? Talk to us.