An app in every chat tool
A Slack app, a Teams bot registration, a Google Chat app. Each with its own manifest, its own event shapes, its own way of threading a reply and its own review process.
Customer facing AI
Ship your AI into your customers' workspaces.
Their staff ask in Slack, Teams or Google Chat. Your assistant answers, under your name, using that customer's own data and permissions. You build the assistant. Everything between it and them is already built.
Meetext holds one box. The model, the prompt and the name on the bot are yours, and the conversation is never stored.
The assistant is the easy half.
Most teams selling AI into enterprises already have one working. It answers well in a demo, against their own data, in an account they control. Then the first customer asks for it inside their Slack, and the work that follows has nothing to do with the model.
That work is the same for every customer after the first, which is exactly the kind of work that should be a dependency rather than a quarter.
What you would otherwise build
Knowing whose workspace this is
A message arrives with a workspace id and nothing else. Turning that into the right customer, with the right credentials, is the part that must never be wrong once, because being wrong means answering one customer from another's data.
Per-customer access to your own product
Your assistant needs to reach your systems as that customer, with that customer's permissions, limited to what they bought. Not as you, and not as everybody.
Consent, tokens and the day they expire
An enterprise admin approves the install. The grant belongs to them, refreshes on their schedule and gets revoked on their timetable. Somebody has to run that lifecycle per customer, forever.
A security review, per customer
Every enterprise buyer asks what your AI can reach inside their systems before it is allowed in. Answering that with a document somebody wrote by hand does not survive the second customer.
Streaming that does not look broken
A model takes twenty seconds. Twenty seconds of silence reads as broken, one message per token is unreadable, and every provider rate limits edits. There is a right answer and it is not obvious.
We are the wire, not the voice.
Meetext carries a question one way and an answer the other. It is deliberately the least interesting part of the exchange, because everything interesting about your product is yours and should stay that way.
The model
Whichever you use. We never see it.
The prompt
Your assistant, your instructions, your behaviour.
The name
The bot answers as you. Meetext is not mentioned.
The conversation
We send a question and a thread reference. Never history.
What your assistant is handed
Not just the words. Every question arrives with that customer's own connection to your product, so your assistant can look things up and act for them through their credentials, limited to the version they installed and the capabilities they are entitled to.
That boundary is not advisory. It is the same per-customer endpoint their own AI tools would use, with the same isolation a security review asks about first.
Built and tested, not yet run against a real workspace
The bridge is complete and covered end to end against a provider simulator, and it can be exercised locally today. It has not carried a question in a real Slack workspace, Teams tenant or Google Chat space, because that needs OAuth applications that are still being registered. Every destination on this site reads Preview for the same reason, and will keep reading Preview until something real has actually run.
Where it can land
Questions arrive from Slack, Microsoft Teams and Google Chat. The deployment behind them models Atlassian, Google Workspace, HubSpot, Microsoft 365, Salesforce, ServiceNow and Slack, so the assistant and the rest of the installation are the same object with the same version and the same entitlements.
Questions
Whose AI is it?
Yours. Meetext posts each question to an endpoint you run and relays what you send back, so the model, the prompt, the tools you add on your side and the name on the bot are all yours. Nothing identifies Meetext to the person asking. If you have not built an assistant yet, Meetext can run a model for you as a starting point, and that one is still given only your capabilities to answer from.
Do you store the conversation?
No. Each question is sent with a stable thread reference and your assistant keeps its own memory against that reference. Meetext does not send you the history and does not keep it, which is deliberate: it is the difference between a claim about retention and a system that has nothing to retain.
How does my assistant get at my product's data for a specific customer?
Every question arrives with that customer's own endpoint and token attached, which is the same pair the customer's own AI tools would use. Your assistant calls your systems through it. What it can reach is bounded by the version that customer installed and by what they are entitled to, so it cannot call something you have not approved and cannot see another customer's data.
What if my assistant is down, or slow, or fails halfway?
The person asking is told something rather than left waiting. A reply that fails partway keeps whatever arrived and adds the reason. Your own error text goes to your event log where you can act on it, and never into your customer's channel, because a stack trace from your server appearing in their Slack is its own incident.
Can I try it before registering apps with Slack and Microsoft?
Yes. There is a Try it control on the product page that sends a question down the real path and shows you the reply instead of posting it into a channel, with the latency, whether the answer actually streamed, and which customer's capabilities it was given. That is how you find a broken endpoint, rather than by watching a customer find it.
Does this replace the MCP endpoint?
No, it sits alongside it. A customer whose staff use their own AI tools connect those to the MCP endpoint. A customer who would rather ask in Slack gets the assistant. Both are the same deployment, the same version and the same entitlements, so nothing has to be decided twice.
2 environments free
Stop assigning an engineer to every customer.
Connect a source, publish, and send one link. Your next enterprise customer installs itself.
2 customer environments free, forever. No card required.