Bourdon plugin for OpenClaw — cross-agent memory federation native to your OpenClaw agents.
After installing this plugin, your OpenClaw agents:
- Publish an
openclaw.l5.yamlmanifest to~/agent-library/agents/after every session, so other Bourdon-equipped agents (Claude Code, Codex, Cursor) can recall what OpenClaw did. - Recall what other agents have been doing via six new tools:
| Tool | Description |
|---|---|
bourdon_search |
Search a specific agent's L5 manifest for entities and recent work matching a topic. |
bourdon_cross_agent_summary |
Summarize a project's state across every agent that's worked on it. |
bourdon_recent_work |
List recent sessions across all agents (or filtered to one) since a date. |
bourdon_find_entity |
Look up an entity (project, file, concept) across agents and return everywhere it's been touched. |
bourdon_recognize |
Format a Bourdon recognition payload for a given prompt — the recognition-first runtime hook. |
bourdon_deeper_context |
Pull deeper episodic context for a prompt when the agent has the response time. |
Plus the agent-library:// MCP resources for direct manifest browsing.
The Bourdon thesis: cross-agent memory federation works when every agent has a Bourdon layer. OpenClaw is a 370k-star personal AI assistant platform with a real plugin SDK, daily-cadence releases, and a culture of AGENTS.md adoption. Wiring Bourdon into OpenClaw means OpenClaw users get cross-agent recall against the rest of their tooling without leaving the assistant.
openclaw plugin install bourdonpnpm add -D @bourdon/openclawThen add to your OpenClaw config (~/.openclaw/config.json or whatever your installation uses):
{
"plugins": {
"entries": {
"bourdon": {
"enabled": true
}
}
}
}- Bourdon installed in a reachable Python environment. The plugin spawns
python -m core.l6_serverto query Bourdon's L6 federation server.pip install bourdon(currently 0.3.0) on the same host where OpenClaw runs. - At least one L5 manifest under
~/agent-library/agents/. The fastest way to seed: enable Bourdon's Claude Code SessionEnd hook and end one Claude Code session. After that, you have something to recall.
Default config (everything optional):
{
"library": "~/agent-library",
"publishOnSessionEnd": true,
"accessLevel": "team",
"l6": {
"transport": "stdio",
"command": "python",
"args": ["-m", "core.l6_server"],
"spawnTimeoutMs": 5000
},
"redaction": {
"snippetCharLimit": 180
}
}For users who want L6 as a long-running HTTP service rather than a per-call stdio subprocess:
{
"l6": {
"transport": "http",
"url": "http://localhost:7500/sse"
}
}Then start L6 separately:
python -m core.l6_server --transport http --port 7500 --library ~/agent-libraryThe plugin defaults to accessLevel: "team". Entities tagged private in their source L5 manifest are filtered at the L6 layer and never reach the OpenClaw agent. Override per-call by passing access_level to any bourdon_* tool.
To prevent OpenClaw from seeing certain agents' memory entirely, scope the library setting to a directory that only contains the manifests you want to expose.
pnpm install
pnpm test
pnpm buildThis is a community plugin built against OpenClaw's published openclaw/plugin-sdk/* subpath exports — no monorepo clone required. If you're contributing back to OpenClaw's extensions/bourdon/ directly, the source files port over cleanly with a workspace-deps adjustment.
- Bourdon project: bourdon.ai · getbourdon/bourdon
- OpenClaw project: openclaw.ai · openclaw/openclaw
- Bourdon adapter authoring guide:
bourdon/docs/AUTHORING_AN_ADAPTER.md - MCP specification: modelcontextprotocol.io
Business Source License 1.1 — auto-converts to Apache 2.0 four years after the version's publication. Same license as Bourdon itself. Free for solo, internal, research, and non-competing commercial use. Commercial license needed only for hosted-service offerings or paid-product embedding that competes with RADLAB LLC's paid version of Bourdon. See Bourdon's LICENSE_FAQ.md for plain-English guidance, or contact licensing@bourdon.ai.