Last updated: 2026-06-04
This privacy policy describes how the agmsg project (the software at https://github.com/fujibee/agmsg, distributed via agmsg.cc, the Anthropic / OpenAI / community plugin and skill marketplaces, and the agmsg npm package) handles user data.
In short: agmsg does not collect, transmit, or share any user data. Everything stays on the user's machine.
When installed, agmsg stores the following on the user's local filesystem only:
- Team and agent identity registry — JSON files under
~/.agents/skills/agmsg/teams/<team>/config.json. Contains team names, agent names within each team, the agent type label (claude-code,codex,gemini,antigravity,copilot,opencode), and project paths the user joined from. The user explicitly chooses every value at join time. - Messages sent between local agents — rows in
~/.agents/skills/agmsg/db/messages.db, an SQLite file. The message body, sender, recipient, team, and timestamp are stored. The user (or the agent acting on their behalf) chooses every value when invokingsend.sh. - Per-session runtime files — pidfiles, last-checked markers, and the actas exclusivity lock files under
~/.agents/skills/agmsg/run/. These reference the user's session IDs and process IDs to coordinate hooks. - Hook configuration in the user's projects — agmsg writes per-runtime hook files (e.g.
<project>/.claude/settings.local.json,<project>/.codex/hooks.json,<project>/.agent/rules/agmsg.md,<project>/.github/hooks/agmsg.json) when the user picks a delivery mode. These contain absolute paths to the agmsg scripts; they do not contain personal data beyond filesystem paths.
All of the above lives on the user's machine. agmsg has no daemon, no server, and no remote endpoint.
- No network requests. The
install.shstep usescurlto fetch the installer itself, and the optionalagmsg.ccredirect goes through GitHub. After installation, the runtime makes no outbound requests. - No telemetry or analytics. No usage data, error reports, or counts are sent anywhere.
- No third-party services. agmsg integrates with whatever CLI AI agent the user has installed (Claude Code, Codex, Gemini CLI, Copilot CLI, Antigravity, OpenCode); it does not call those agents' backends itself. Anything the user types to one of those agents is governed by that agent's own privacy policy, not by agmsg.
- No accounts. agmsg has no user accounts, no login, no API keys.
- No data sharing. Because agmsg never receives data from a remote endpoint and never sends data to one, there is nothing to share, sell, or disclose.
All data agmsg writes is on the user's local filesystem under:
~/.agents/skills/agmsg/(the skill directory)<project>/.claude/,<project>/.codex/,<project>/.agent/,<project>/.github/hooks/(per-project hook configs, in the user's own project directories)
The user is solely in control. To delete everything agmsg has stored, run uninstall.sh from the repo (which removes the skill directory and per-project hook files) and then delete ~/.agents/skills/agmsg/ if anything remains.
When two agents on the same machine use agmsg to communicate, the message body is written to messages.db by the sender and read back by the recipient when their hook fires. The bytes never leave the user's filesystem. Whether the content of those messages is sensitive is up to the user — agmsg treats the body as opaque text.
agmsg does not collect data from anyone, including children. The software is a developer tool intended for use by users 13 and older in accordance with the underlying agent CLIs (Claude Code, Codex, etc.) and their respective terms.
If agmsg ever begins to collect or transmit data, this policy will be updated and the change announced in the project repository's CHANGELOG or release notes. The current version of this policy lives at:
https://github.com/fujibee/agmsg/blob/main/PRIVACY.md
Questions or concerns about this policy can be raised by:
- Opening an issue at https://github.com/fujibee/agmsg/issues, or
- Emailing the maintainer at fujibee@gmail.com.
The agmsg project itself is MIT-licensed. See LICENSE.