I build Brief, a Chrome extension for reporting bugs. You click it, show the bug while you talk, and hit stop. Brief writes the ticket, with steps to reproduce, screenshots, console errors and failed requests, and files it to Linear, Jira, GitHub or seven other trackers.

Teams liked it. Then they asked the question that stopped almost every deal:

"Our bug videos show customer data. How do we know you don't look at them?"

The usual answer costs money I don't have

Big vendors answer with certificates. ISO 27001, SOC 2, a security page full of logos. Getting there takes months and tens of thousands of euros in audits and tooling. I'm one person. "Trust me" was all I had, and it isn't an answer.

So I tried the opposite of a certificate: make every claim something you can check yourself.

1. The admin views can't read your tickets

Every SaaS has an internal admin page. Mine used to show everything, ticket titles included. Now it reads a fixed list of database columns, and none of them hold ticket content:

export const OWNER_COLUMNS = [ "id", "user_id", "status", "created_at", "archived_at", "ticket_provider", "filing_started_at", ] as const; 

A test fails if anyone adds a column to that list. So "we can't see your tickets" isn't a policy, it's a build that breaks.

2. You can see exactly what we see

Every user has a What we see page. It shows their own row of the admin view, drawn by the same code the admin page uses. Not a description of the data, the data.

3. Recordings don't stay

Once a ticket is filed, the audio track and the video frames used to write it are deleted. What stays is the ticket itself, so your feed can show it.

4. The whole thing is open source

Brief is now MIT licensed: https://github.com/razi2001/brief

Open code is only reassuring if someone reads it, and nobody has time to audit a stranger's repo. So I added a file, docs/AUDIT.md, written for a coding agent. You paste one line into Claude Code, Cursor or any agent:

Clone https://github.com/razi2001/brief and audit it by following docs/AUDIT.md. 

It checks each privacy claim against the code (which columns the admin views read, what the activity log stores, what leaves a self-hosted copy, what gets deleted, row level security on every table), runs the guard tests, and reports every place where the code does more than the privacy page says.

I'd genuinely like to see what your agent finds.

5. Or don't trust me at all

You can run Brief on your own servers with docker compose up and your own Supabase, Clerk and OpenAI keys. There's a setup prompt for that too, in docs/SELF_HOSTING.md. Then nothing ever reaches me.

What I learned

A small team can't buy trust. But it can make trust unnecessary: limit what the operator can see in code, show users that limit, and hand them a tool to verify it. "Don't trust me, check it" turned out to be a better pitch than any logo I could have bought.

If this sounds useful, try it at https://get-brief.app, read the code at https://github.com/razi2001/brief, and if you like the idea, a star helps a lot.