Coding-agent sync and transcripts

Your coding agents already write a log of every session. Living Brain can read those logs, so the decisions and dead ends in them become searchable and citable. Nothing is read or sent from a repository you have not opted in.

What is read

• Claude Code sessions, under ~/.claude/projects.
• Codex sessions, under ~/.codex/sessions.

Opt in per repository

Only repositories you add are synced. In the Mac app, choose Add Project… and pick a folder; from the terminal, run livingbrain logs allow <path>. A session recorded in any other repository is reported as skipped, and nothing about it is sent.

When it syncs

• Live sync (on by default in the Mac app): a session syncs once it has had no new activity for two minutes.
• Sync Now runs a sync at once.
• Sync Automatically Every 15 Min is an option you can turn on.
• From a terminal: livingbrain logs sync.

Pause Syncing stops live and scheduled sync. Sync Now still works.

What is redacted

Secrets and email addresses are removed on your machine, before anything is sent. The server runs the same redaction again. Redaction catches the common patterns; it cannot know every secret, so check the preview in the Mac app before you rely on a repository being clean.

Transcripts

Open Sources → Agent sessions in the web app, or a session in the Mac app, to read the whole session: your messages, the agent's replies, its reasoning summaries, each tool call with its command or patch, and the clipped output. The Mac app shows the exact redacted record that would be sent before you sync it.

Subagent sessions are nested under the tool call that started them, so a delegated task reads in place.

The compiled wiki

A compile step reads the synced sessions one at a time and maintains Markdown pages: one index and patterns page per agent, and pages per project. Every claim cites its session. Compiles run in the background on a budget you can set under Settings → Usage & limits, and the web app shows the queue and when the next run is due. The full design is in the coding-agents reference.

Use your own subscription for brain jobs

The compile and the answers to questions are model jobs. On your Mac you can run them with your own Claude Code or Codex subscription instead of the workspace's model, from Run Brain Jobs with My Subscription in the Mac app, or with livingbrain jobs run on a headless machine. The agent runs with tools off and a read-only sandbox, and Living Brain never holds your agent credentials. Details are in the local runner reference.

Related

• Agent logs: the logs commands in full.
• Getting started: sign in and install the Mac app.