OpenClaw

OpenClaw should use pre-turn recall and post-turn async writeback with scoped identity. Handoffs should be explicit summaries with allowed memory references, not transcript inheritance.

Registry contract

This section is generated from integrations/agent_integrations.registry.json. Edit the registry or generator, not this block.

Field	Value
Integration ID	openclaw
Label	OpenClaw
Tier	native_lifecycle_provider
Setup mode	plugin
Agent ID	openclaw
App ID	openclaw
Source app / actor	source_app=openclaw, agent_id=openclaw; key metadata wins over request body hints
Default scope	own
Config paths	~/.openclaw/openclaw.json, ~/.openclaw/.env
Capabilities	memory.read, memory.write, memory.recall, memory.delete, agent.context, agent.writeback, mesh.read, evidence.read, courtroom.read, canary.read
Supported tools	um_store, um_search, um_forget, um_profile, um_courtroom, unifiedmemory_store, unifiedmemory_search, unifiedmemory_profile

Capture policy

before_prompt_build injects governed recall; agent_end capture writes through writeback admission with loop-prevention metadata.

Install command(s)

openclaw plugins install ./integrations/openclaw-unifiedmemory --link
openclaw unifiedmemory setup --write --api-url "{{API_V1_BASE_URL}}" --api-key-env UNIFIEDMEMORY_OPENCLAW_API_KEY --container-tag "{{CONTAINER_TAG}}" --long-term-container "{{CONTAINER_TAG}}" --session-container "{{CONTAINER_TAG}}" --recall-scope long-term --request-timeout-ms 30000

Verify command(s)

openclaw config validate
openclaw plugins doctor
openclaw plugins inspect openclaw-unifiedmemory --runtime --json
openclaw unifiedmemory status
openclaw unifiedmemory search --query "UnifiedMemory setup" --limit 3

Environment

{
  "UNIFIEDMEMORY_OPENCLAW_AGENT_SCOPE": "own",
  "UNIFIEDMEMORY_OPENCLAW_API_KEY": "{{KEY_ENV_REF}}",
  "UNIFIEDMEMORY_OPENCLAW_API_URL": "{{API_V1_BASE_URL}}",
  "UNIFIEDMEMORY_OPENCLAW_AUTO_CAPTURE": "true",
  "UNIFIEDMEMORY_OPENCLAW_AUTO_RECALL": "true",
  "UNIFIEDMEMORY_OPENCLAW_CONTAINER": "{{CONTAINER_TAG}}",
  "UNIFIEDMEMORY_OPENCLAW_LONG_TERM_CONTAINER": "{{CONTAINER_TAG}}",
  "UNIFIEDMEMORY_OPENCLAW_RECALL_SCOPE": "long-term",
  "UNIFIEDMEMORY_OPENCLAW_SESSION_CONTAINER": "{{CONTAINER_TAG}}"
}

Risk notes

• Do not set plugins.slots.memory to openclaw-unifiedmemory.
• Use the same canonical workspace for session and long-term containers unless an admin explicitly grants a separate project workspace.

Official upstream docs

https://docs.openclaw.ai/plugins/hooks

Scope and capabilities

Setting	Value
Source app	openclaw
Default container	jeethendra
Default scope	own
Key env var	UNIFIEDMEMORY_OPENCLAW_API_KEY

Install

openclaw plugins install ./integrations/openclaw-unifiedmemory --link
openclaw unifiedmemory setup --write \
  --api-url "https://edge-api.jithendranara.dev/api/v1" \
  --api-key-env UNIFIEDMEMORY_OPENCLAW_API_KEY \
  --container-tag "jeethendra" \
  --long-term-container "jeethendra" \
  --session-container "jeethendra" \
  --recall-scope long-term \
  --request-timeout-ms 30000

Verify

Use the plugin verification command from the setup output. It should check the selected scoped key, not an owner browser session.

Expected proof:

• recall returns own memory and shared canonical owner facts;
• writeback can save a disposable outcome;
• forget/delete works only where advertised by key capabilities;
• Courtroom read works;
• cross-agent private memory does not leak.

Handoff rules

OpenClaw-to-agent handoffs should include:

• handoff_from;
• handoff_to;
• a short summary;
• explicit allowed memory references.

Do not pass entire transcripts as inherited memory.