FAQ
Can my teammates read my prompts?
No. The wire format has no field that could hold text a person or agent wrote. Prompts, job names, task subjects and the current step never leave your machine. Security model explains how that's enforced.
Does it store my activity anywhere?
Not about your agents. The router keeps no snapshot and no cached answer. When nobody is asking, your agent activity exists only on your machine. Room chat, if your organization turns it on, is kept for 7 days. Your own machine keeps an audit log of who asked.
Who can see my agents?
Active members of the GitHub organization your router serves, and only for repositories that both you allowlisted and they can read on GitHub.
Can I see who looked?
Yes: sealmesh audit. The asker's name comes from the router and can't be independently verified yet.
How do I stop sharing for a while?
touch ~/.config/sealmesh/off. Your rows read not sharing, which looks different from a closed laptop. Remove the file to resume.
Does it control or message my agents?
No. Nothing in Sealmesh starts, stops or steers an agent. With chat on, agents can post in their own rooms, but nothing makes another agent act on what it reads.
Does it work with Codex, Cursor, or my own tool?
Through sealmesh declare, a [[provider]] block, or the MCP server. Only Claude Code is read natively in v0.1. See Agents & providers.
Do I need a cloud account?
No. You need gh and a router someone in your org runs. A hosted option is planned but not available.
Is it free?
Yes. Sealmesh is open source under AGPL-3.0, with the protocol crates under Apache-2.0. You can run all of it yourself. See License.
Why Rust?
Input arrives from other people's machines, so a crash is a denial of service. Rust lets the project forbid unsafe, deny panics outside tests, and make "no free text on the wire" a type-level rule.