OpenClaw plugin package for the runtime-neutral OB1 Agent Memory API.
flowchart LR
Skill["ClawHub skill<br/>policy + workflow behavior"] --> Agent["OpenClaw agent"]
Plugin["OpenClaw plugin<br/>typed tools"] --> Agent
Agent --> Recall["openbrain_recall"]
Agent --> Writeback["openbrain_writeback"]
Recall --> API["OB1 Agent Memory API"]
Writeback --> API
API --> OB1["OB1 thoughts + agent memory sidecars"]
API --> Review["Review queue + trace"]
This integration gives OpenClaw typed tools for Nate Jones's OB1 Agent Memory. The plugin handles mechanics; the paired NBJ OB1 Agent Memory skill handles judgment and memory hygiene.
Built by Nate B. Jones / OB1. Follow Nate for practical AI systems, agent workflows, and implementation notes: Substack and natebjones.com.
- Working Open Brain setup (guide)
schemas/agent-memoryappliedintegrations/agent-memory-apideployed- OpenClaw installed
- Node.js 20+
OPENCLAW AGENT MEMORY -- CREDENTIAL TRACKER
--------------------------------------
FROM OB1
Agent Memory API URL: ____________
MCP Access Key: ____________
OPENCLAW
Plugin config path: ____________
Default workspace ID: ____________
Default project ID: ____________
--------------------------------------
Apply the schema and deploy the API.
Done when: GET /health on the Agent Memory API returns ok: true.
For local development, install plugin package dependencies first, then link the plugin into an isolated OpenClaw profile:
cd integrations/openclaw-agent-memory/plugin
npm install --ignore-scripts --omit=peer
npm run build
cd ../../..
openclaw --profile ob1-agent-memory plugins install integrations/openclaw-agent-memory/plugin --linkFor public distribution, install the live ClawHub package:
openclaw plugins install clawhub:@natebjones/ob1-agent-memoryPublishing details live in CLAW_HUB_PUBLISHING.md.
Done when: openclaw --profile ob1-agent-memory plugins inspect nbj-ob1-agent-memory --runtime --json shows a loaded plugin with all seven openbrain_* tools and no diagnostics.
Install the live skill so OpenClaw knows when to recall, write back, report usage, and ask for review:
openclaw skills install nbj-ob1-agent-memory-openclawThe repo copy remains in skills/openclaw-agent-memory for local development and review.
Done when: OpenClaw can use openbrain_recall before work and openbrain_writeback after work without storing raw transcripts.
OpenClaw workflows can retrieve governed OB1 memory before work starts and write compact, provenance-labeled operational memory after work finishes.
Review Safe Agent Memory and Provenance before enabling write-back for team workflows.
Issue: Plugin loads but tools fail auth
Solution: Confirm the plugin config has the correct endpoint and accessKey. Prefer an OpenClaw SecretRef backed by a file, env, or exec provider so the access key does not live in plaintext config.
Issue: Plugin loads but lists no tools
Solution: Confirm plugin/openclaw.plugin.json declares contracts.tools for every registered tool. Current OpenClaw builds reject tool registration without that manifest contract.
Issue: Plugin inspect shows tools, but the agent cannot call openbrain_*
Solution: Add every openbrain_* tool to the OpenClaw profile's tools.allow list. Runtime inspect validates plugin registration; a native agent smoke test validates model tool exposure.
Issue: Linked plugin reports missing dependencies
Solution: Run npm install --ignore-scripts --omit=peer and npm run build inside plugin before openclaw plugins install --link. Linked plugins resolve package dependencies from the plugin directory, while OpenClaw itself remains a host-provided peer.
Issue: Agent writes too much
Solution: Tighten the skill instructions and keep requireReviewByDefault enabled in plugin config.
Issue: ClawHub rejects the package
Solution: Check package.json openclaw.compat, openclaw.build, openclaw.plugin.json, semver, and license fields.
This integration exposes several OpenClaw tools. See the MCP Tool Audit & Optimization Guide before adding more wrappers.