← Docs hub

OpenClaw adapter

Reads .jsonl session transcripts written by the OpenClaw agent gateway (native store under ~/.openclaw/agents/, not the OpenCode app-config layout — that is the separate opencode adapter).

Runs on bare llmwiki sync when configured roots exist (native ~/.openclaw/agents/ or vault inbox at <vault>/.openclaw-sessions-inbox).

Session store

Default root: ~/.openclaw/agents/ (same path on Linux, macOS, and Windows via Path.home()).

Layouts accepted under each configured root:

• Native install: <agent>/sessions/<uuid>.jsonl
• Vault inbox mirror: <agent>/<uuid>.jsonl (no sessions/ segment) — used when another host copies transcripts into a Syncthing vault at e.g. <vault>/.openclaw-sessions-inbox/

Skipped (not conversation transcripts): *.trajectory.jsonl, *.checkpoint.*.jsonl, and anything under a _quarantine/ directory.

Override the search roots in config.json:

{
  "adapters": {
    "openclaw": {
      "roots": ["~/.openclaw/agents", "<vault>/.openclaw-sessions-inbox"]
    }
  }
}

What it reads

Each session is a JSONL stream of typed OpenClaw records. Only type == "message" rows become conversation turns; user content lists are flattened to strings for the shared renderer.

Automated (headless) sessions

All OpenClaw sessions are treated as not headless. filters.exclude_headless does not skip them. See Multi-agent setup — What “automated” means.

Enable it

python3 -m llmwiki sync --adapter openclaw

Point roots at a vault inbox when transcripts are mirrored there instead of ~/.openclaw/agents.

Output layout

Standard raw/sessions/<YYYY-MM-DDTHH-MM>-openclaw-<agent>-<slug>.md (project slug is openclaw-<agent>).

Code

• llmwiki/adapters/contrib/openclaw.py
• Tests: tests/test_openclaw_cursor_cli_adapters.py

See also

• OpenCode adapter — shared schema under OpenCode / OpenClaw app-config dirs
• Multi-agent setup
• Configuration reference — adapters.openclaw.roots