--- title: '`glab minder setup`' stage: AI Clients group: Developer Clients info: To determine the technical writer assigned to the Stage/Group associated with this page, see --- Configure this machine to record AI agent sessions. (EXPERIMENTAL) ## Synopsis Configure the current machine for AI governance session capture. Choose the agents to sync with `--agents`. Without it, setup syncs the agents set up by an earlier run, or Claude Code if there are none. For each agent, setup installs: - Claude Code (`claude-code`): a Stop hook in `~/.claude/settings.json`, which syncs the session after every agent turn, and a SessionEnd hook, which syncs the session and marks it completed. - OpenCode (`opencode`), version 1.4 or later: a plugin in `~/.config/opencode/plugins/glab-govern.js` (under `$XDG_CONFIG_HOME` when it is set), which syncs the session every time it goes idle. OpenCode loads it for all your sessions, in every project. Later setups keep it up to date. - Codex (`codex`) and Cursor (`cursor`): nothing. The fallback periodic sync finds their sessions by scanning their transcripts in `~/.codex/sessions` and `~/.cursor/projects`, and uploads those from repositories on GitLab hosts you are logged in to with glab. Their sessions are marked completed once they have been idle for 24 hours. Setup also installs the fallback periodic sync: a launchd agent (`~/Library/LaunchAgents/com.gitlab.glab-govern-audit-sync.plist`) on macOS, or a systemd user timer (`~/.config/systemd/user/glab-govern-audit-sync.timer`) on Linux. The fallback periodic sync runs `glab minder audit sync --all` every 30 minutes, and on Linux also 5 minutes after boot, until you remove it. It uploads anything the hooks missed for sessions they have already recorded (see `glab minder audit sync --help` for what is uploaded), using your stored glab credentials and the glab configuration directory in use when you run setup. Each session goes to the project and host of the Git repository the agent ran in. It also marks sessions completed once they have been idle for 24 hours. It pauses sessions from agents that use hooks while those hooks are not installed. Pass `--no-fallback-sync` to install only the hooks and plugin, which Codex and Cursor cannot be synced without. The fallback periodic sync is not available on Windows. By default, glab uploads only what each agent does after you first set it up: not its sessions from before, and not the earlier turns of a session from before that you resume. Cursor records no times, so setup notes where each existing Cursor transcript ends, and only what is added to it afterwards is uploaded. A chat that Cursor copies under a new ID, for example when you duplicate it, counts as a new session and is uploaded in full. Pass `--include-history` to also upload activity from the last 89 days, including those sessions and turns, for the agents this setup sets up for the first time. For Cursor, that is every transcript changed in the last 89 days. An agent set up earlier keeps what it uploads, so to add its history, run `--uninstall` first. If you set up an agent with an earlier glab version, which did not record when, its activity from the last 89 days is still uploaded until you run setup again. Run setup again with `--agents` to add agents. Agents set up by an earlier run stay set up, so to stop syncing an agent, run `--uninstall` and then set up the agents you want. Every setup prompt names the agents that are synced. Safe to run multiple times: existing hooks are not duplicated, and an existing fallback periodic sync job is replaced. Run `glab minder doctor` afterwards to verify the setup. Run `glab minder setup --uninstall` to remove everything setup installed: the hooks, the OpenCode plugin, the fallback periodic sync job, and Codex and Cursor session syncing. It also removes glab's local records of your agent sessions and forgets the agents you set up and when, so the next setup starts with Claude Code again. Other hooks and settings in `~/.claude/settings.json`, and other OpenCode plugins, are left unchanged. This feature is an experiment and is not ready for production use. It might be unstable or removed at any time. For more information, see . ```plaintext glab minder setup [flags] ``` ## Examples ```console # Sync Claude Code sessions, or the agents set up earlier $ glab minder setup # Sync OpenCode sessions $ glab minder setup --agents opencode # Sync Claude Code, OpenCode, and Codex sessions $ glab minder setup --agents claude-code,opencode,codex # Also upload activity from the last 89 days, including sessions from before setup $ glab minder setup --include-history # Install only the hooks and plugin, without the fallback periodic sync $ glab minder setup --no-fallback-sync # Remove everything setup installed $ glab minder setup --uninstall ``` ## Options ```plaintext --agents strings Agents to sync sessions from: claude-code, opencode, codex, cursor. Added to the agents set up by an earlier run. The OpenCode plugin runs for all your OpenCode sessions. Multiple agents can be comma-separated or specified by repeating the flag. --include-history Also upload the activity from the last 89 days of the agents set up for the first time, including sessions from before setup and the earlier turns of sessions you resume. --no-fallback-sync Install only the hooks and plugin, without the fallback periodic sync job. --uninstall Remove the hooks, the OpenCode plugin, the fallback periodic sync job, and the local session records, and stop syncing Codex and Cursor sessions. -y, --yes Skip the confirmation prompt, and agree to what it would describe: the changes and, for setup, the uploads. ``` ## Options inherited from parent commands ```plaintext -h, --help Show help for this command. ```