The AI never sees them
Our servers hold your keys and use them to fetch data. The AI only receives the data that comes back — never the key, in the question, the answer, or anywhere in between.
Connecting your analytics, billing and app data to anything means trusting it with passwords and keys. So here's the plain version of how we handle yours — what's protected, what the AI can and can't touch, and how to switch it off the moment you want to.
Read-only by default. You can disconnect any source in one click.
Your question
"Why did day-three return drop last week?"
The numbers your tools sent back
Retention, sessions, revenue — the data itself.
Your API keys and passwords
Never. Not once, not in any form.
Our servers use your keys to fetch the data. Only the data goes to the AI.
Four promises
The AI never sees them
Our servers hold your keys and use them to fetch data. The AI only receives the data that comes back — never the key, in the question, the answer, or anywhere in between.
Locked, not lying around
Every key is encrypted with AES-256-GCM, and the code that unlocks it is kept somewhere separate. Even someone holding a copy of our database would have nothing usable.
Off the moment you say so
Disconnect a source and it stops working immediately — there's no copy sitting in a cache somewhere to keep going. Delete it and the stored key is destroyed outright.
Read-only by default
Read-only by default. Any source that can write says so when you connect it. A read-only key can look at your data but cannot change or delete a single thing — so even a worst case stays a worst case you can live with.
How a question actually works
Step one
The AI works out which of your tools it needs to look at. It knows the tool's name and what to ask it — nothing more.
Step two · our servers
Your key is unlocked, used for that one request, and forgotten again. This all happens on our side — the AI isn't part of it.
Step three
It gets the data back and explains what it means. Anything that looks like a password or key is stripped out before it arrives, as a second line of defence.
Put a key of yours into an AI prompt
Store a key as plain readable text, anywhere
Write one into a log file or an audit record
Let the AI change, delete or spend anything on its own
Let another customer's account reach your data
Which tools you connect, one at a time
How much access each one gets — read-only by default
Disconnecting any of them, instantly, without asking us
Seeing a full record of every connection change
Rotating the key at the source, any time you like
Straight answers
It can't leak what it never had. The key is used before the AI is involved, and a second filter strips anything key-shaped out of what it does receive. Two separate things would have to fail at once.
Not unless you deliberately turn on an action — like posting a report to Slack or creating a task. Everything else is view-only by design, and write-capable sources say so when you connect them.
Disconnect a source and access stops on the spot. For total peace of mind you can also rotate the key in the tool itself — that's the switch we can never touch, and it's always yours.
No. Each customer's connections are isolated by tenant in the database and in every query path, so there's no path from someone else's account to yours.
Yes — every connect, test, disconnect and delete is logged with who did it and when. The record never contains the key itself, only what happened to it.
Not yet, and we'd rather say so than imply otherwise. What exists today is the architecture on this page, which we're happy to walk through in as much detail as you want.
Still have a question? Ask us directly — we'd rather answer something specific than have you take our word for any of this.
You can disconnect it in one click. Free while we're in early access.