Skip to main content

​
Session Journal

When an agent session resets via /new or /reset, WednesdayAI can write a reflective journal entry capturing what was discussed — useful for ongoing conversational workflows where a written record matters. Journal entries are Markdown files written to the agent’s workspace. In multi-user mode they are isolated per user lane, so each user’s journal is private.

​
Who it is for

Session journal is designed for agents that hold ongoing conversations across sessions: coaching, patient-facing assistants, customer support, educational tutors, or any scenario where the operator needs a durable record of each interaction.

​
Enabling session journal

Two steps are required: Step 1 — Enable the hook:
Step 2 — Opt the agent in via openclaw.json:
Both steps are required. The hook must be enabled at the gateway level, and the agent must opt in individually — other agents on the same gateway are not affected.

​
Configuration

​
Custom prompt example

Include {transcript} in your prompt where the session content should be inserted:

​
What the output looks like

Journal files are written to <workspaceDir>/journal/ with the name YYYY-MM-DD-HHMMss-<slug>.md, where the slug is derived from the first heading in the LLM-generated text. Example:
Each file contains a header block (agent ID, session key, date/time) followed by the LLM-generated journal body. When workspace lanes are active (multi-user mode), each lane gets its own journal/ folder inside the lane’s workspace directory — journals are isolated per user and not shared.

​
Limitations

Journal entries are written only when a session explicitly resets via /new or /reset. Sessions that end without a reset command (process exit, connection drop, idle timeout) do not produce a journal entry. The /new and /reset commands are held until journal generation completes. With a remote provider and a long transcript, this can take up to ~15 seconds. If latency is a concern, disable the journal for that agent or use a faster provider.

​
Operator notes

​
Disk usage

One Markdown file is written per session reset per opted-in agent. File sizes depend on session length and the LLM’s output verbosity — typically 1–10 KB per entry. Operators should plan for routine cleanup or archiving of journal/ directories in long-running deployments.

​
LLM call cost

Journal generation makes one completion call per session reset. This call uses the configured provider and model for that agent. If the session has no transcript content (empty session), no LLM call is made — a header-only stub is written instead.

​
Failure behavior

If the LLM call fails (network error, provider error, timeout), the hook writes a header-only stub so the reset event is still recorded. No exception propagates to the user’s session reset flow.

​
Disabling

To disable journal generation for a specific agent, set sessionJournal.enabled: false (or remove the sessionJournal block). To disable the hook entirely across all agents:
See Hooks: session-journal for full technical details.