Getting started
Half an hour, one channel, and something real connected.
No migration, no data import, and no call to book. Sign up, add it to Slack or Microsoft Teams, invite it to one channel and connect one service — the whole thing fits in a lunch break.
The first half hour, at a glance
about thirty minutes, none of it a project
The steps
Five steps. None of them needs a call.
2 minutes
Create an account. The first one founds your organisation.
Sign up on the hosted service with an email address or with Slack. The first sign-up founds your organisation’s account, on the free plan, and everybody after arrives by invitation from the console. No card, and nothing to install on your side.
Or run it yourself: the same binary, MIT-licensed, one container and a public HTTPS address. The self-hosted page is that runbook; the rest of this one is the same either way.
10 minutes
Add it to Slack or Microsoft Teams, and invite it to one channel.
The console’s setup screen hands you to Slack’s own consent screen; a workspace admin approves it. Then invite the bot to a single channel — the noisiest one you have, ideally, because that is where it either earns its place or does not. The setup screen watches for the invite and moves on by itself.
On Microsoft Teams, the same screen offers Teams beside Slack: a Teams admin uploads the app once, you add it to one team, and a code sent to the bot from inside Teams ties the organisation to your account. The Teams steps are on the install page.
Self-hosting? Set ALLOWED_EMAIL_DOMAINS at the same time and guests, Slack Connect members and shared channels are out from the first minute rather than after the first incident.
15 minutes
Connect one service and point it at some documents.
Pick something real — the repository you are on call for, the ticket tracker, the error tracker. A preset fills in the hosts and the auth style, Test connection proves the token before you save, and if you would rather not hold the secret yourself, a setup link sends that job to the person who does.
Drop in whatever documentation you have. It is chunked, embedded and searchable in a couple of minutes, and a document can be scoped so it only answers in the channel it belongs to.
the rest of the half hour
Try to break it.
Ask it something that needs three lookups. Ask it to change something and watch it stop. Feed it a document with an instruction hidden in it. Ask it about a private channel you are not in.
This is the part that decides it. Find the edge on a Tuesday, in a channel that matters, before anyone else does — the guardrails page says what you should see.
whenever you like
Widen it one channel at a time.
Every channel is its own scope: its own instructions, its own model, its own budget, its own connections. Nothing is workspace-wide until you attach it to the workspace.
So the second channel is a decision you make with a month of evidence behind it — not a migration you committed to on day one. When the free plan’s budget is the limit, the size picker is a click away in the console, under Settings → Billing.
What you need
Two things, and neither of them is a project.
A Slack or Teams admin
For ten minutes: in Slack, to approve the app and invite the bot to one channel; in Microsoft Teams, to upload the app once.
One credential
For one service you actually use. Or the name of the person who holds it — a setup link does the rest.
Nothing else, on the free plan
The default open-weight model is included, metered against $5 a month. Self-hosting, bring an OpenRouter key or any OpenAI-compatible endpoint.
No card and no conversation is required for the free plan, and turning it off afterwards leaves your Slack or Teams exactly as it was.
Questions
Asked in most first half hours.
Anything not here, ask on support— or read the guardrails, which answer most of the rest.
Do we need to open a port or expose a URL?
Not on the hosted service — the origin is ours. Self-hosting, the service needs a public HTTPS endpoint so Slack or Microsoft Teams can deliver messages and button presses; a quick tunnel needs no account, and every request is verified before it is processed — Slack’s with the app’s signing secret, Teams’ against the Bot Framework’s signature.
Where does our data go?
On the hosted service it stays in the service’s database, isolated per organisation, and the only thing that necessarily leaves is the model call. Self-hosted, all of it stays on your infrastructure, and the model call goes to the endpoint you chose. Either way, credentials are sealed at rest and injected by the proxy at the network edge, so they are never in a prompt.
Which model do we have to use?
Any OpenAI-compatible endpoint. It ships pointed at google/gemini-flash through OpenRouter, and the workspace, a channel and a single thread can each be switched to a different model from the console.
Can it change things in our systems on the free plan?
Only after a person confirms. Any write is held in the thread with a Confirm card, and writes that grant access go to an approver instead of the asker. Reads are never held. You can also leave the first channel read-only by giving it a connection with no write methods at all.
What does the first half hour cost?
Nothing. The free plan includes $5 of model spend a month, metered and shown under each reply, and the bot pauses at the ceiling rather than billing past it. Pro is a flat fee by plan size plus prepaid AI credit, never per seat, and it is bought in the console with a card — see pricing. Self-hosted, the only cost is your own model calls.
What if we decide against it?
Remove the app — from the Slack workspace, or from Manage apps in the Teams admin centre. There is no data import to unwind and no separate place your team was sent to, so your Slack or Teams is exactly as it was. Self-hosted, stop the container.
Pick a channel and start.
One admin, one credential, and a free plan that needs no card.