Sitin
ProductUse casesPricingDownloadDocsChangelog
Sign inGet startedOpen Sitin
ProductUse casesPricingDownloadDocsChangelogSign in
Get started free
English/简体中文
Sitin

From said to done.

Product

  • How it works
  • Use cases
  • Pricing
  • Download

Resources

  • Docs
  • Changelog
  • Blog

Company

  • About
  • Contact

Legal

  • Privacy
  • Terms
  • Security
© 2026 SitinZoom, Google Meet, Microsoft Teams, Claude, ChatGPT and Codex are trademarks of their owners.
English/简体中文

Security

How Sitin is built - where AI runs, where meeting audio goes, and what our servers hold

2026/11/12

This page describes how Sitin is built. It states facts about the architecture. It is not a certification, and we do not claim any security certification today.

Where AI runs

  • Sitin's servers run no AI. They store your team's data, handle sign-in, keep everyone's screen up to date, and relay live meeting media. They never call an AI model.
  • Every model call happens on a member's own computer. Agent work is done by the Claude Code or Codex already installed and signed in on that computer. Captions, translation and summaries use the model that member configured.
  • Model keys stay on that computer. Keys for model providers, and your Claude Code and Codex sign-ins, are never sent to Sitin.

Where meeting audio goes

Speech is turned into text on the computer of the member whose Sitin joined the meeting, with the speech model that member chose:

  • With a model that runs on the same computer, the audio is processed there and never leaves it.
  • With a model provider the member chose, the audio goes from their computer directly to that provider, under the member's own agreement with that provider.

In both cases Sitin's servers do not process the audio with AI. From this step they receive text only: the transcript, what was settled, and the results.

In meetings hosted by Sitin, audio and video travel between participants through Sitin's real-time media servers, as in any video meeting. That media is relayed, not kept, unless a participant starts a cloud recording.

When your agents attend a meeting on another platform, Sitin joins from your computer as one extra participant. That meeting's media is handled by the platform that hosts it and by your computer.

What an agent can do

  • A new agent is read-only.
  • An agent that you allow to change code works in its own copy and its own branch, and opens a pull request. It never pushes to your main branch.
  • What an agent produces waits for a person to approve it: documents and emails in the inbox, code as a pull request.
  • An agent reads files on its owner's computer. Only what it writes back as a result reaches Sitin's servers.

What our servers hold

  • Account data, and the content your team creates in Sitin: projects, issues, meetings, transcripts, messages, documents, boards and attached files.
  • Recordings, only when a participant starts a cloud recording.
  • For each connected computer: its name, platform, sitin version, and which AI command-line tools are installed and signed in. Not their credentials.
  • Access tokens for services you connect, such as a calendar or GitHub. They are stored encrypted on our servers and are never sent to members' computers.

Each team's data is kept apart from every other team's: every record belongs to one workspace, and every request is limited to the workspace of the person who makes it.

The Privacy Policy lists what we collect and how we use it.

Reporting a problem

If you find a security problem in Sitin, write to hello@sitin.team. Please give us time to fix it before you make it public.