@evermemos/openclaw-evermemos

EverMemOS memory plugin for OpenClaw.

This plugin provides:

• Auto-Recall: retrieve relevant long-term memories before each agent turn and inject them via prependContext.
• Auto-Capture: capture conversation turns after each run and store them in EverMemOS.
• Explicit memory tools: search, store, and profile access from the agent.
• CLI and slash commands for setup/status/search/wipe workflows.

Install

openclaw plugins install @evermemos/openclaw-evermemos

Set memory slot:

plugins:
  slots:
    memory: openclaw-evermemos

Quick Config

plugins:
  entries:
    openclaw-evermemos:
      enabled: true
      config:
        baseUrl: http://localhost:1995
        userId: your-user-id
        recallEnabled: true
        captureEnabled: true

Configuration

Field	Type	Default	Notes
baseUrl	string	http://localhost:1995	EverMemOS server base URL
apiKey	string?	undefined	Reserved for hosted mode
userId	string	default-user	Memory scope identifier
recallEnabled	boolean	true	Enable auto-recall hook
recallMaxResults	number	6	Max injected memory count
retrieveMethod	hybrid | embedding | bm25 | agentic | rrf	hybrid	embedding->vector, bm25->keyword mapping
memoryTypes	string[]	["episodic_memory", "profile"]	Memory types to search
includeProfile	boolean	true	Include profile in context
profileFrequency	number	10	Inject full profile every N turns
captureEnabled	boolean	true	Enable auto-capture hook
captureStrategy	last_turn | all | user_only	last_turn	Turn selection strategy
asyncMode	boolean	true	Capture in fire-and-forget mode
debug	boolean	false	Extra debug logs
queryPrefix	string?	undefined	Optional recall query prefix
tags	string[]	[]	Optional tags used in capture
timeoutMs	number	5000	HTTP timeout for all API calls

Hook Behavior

Auto-Recall

Hooks: before_prompt_build and before_agent_start (compatibility).

Flow:

1. Extract latest user message from current run context.
2. Search EverMemOS (/api/v1/memories/search).
3. Format memory block into:

<evermemos_context>...</evermemos_context>

4. Return { prependContext }.

Failure behavior: logs warning and continues without memory injection.

Auto-Capture

Hook: agent_end.

Flow:

1. Extract turns from run messages.
2. Filter by captureStrategy.
3. Store selected turns via /api/v1/memories.

Failure behavior: never breaks user response path.

Agent Tools

• evermemos_memory_search
  • Query long-term memory.
• evermemos_memory_store
  • Explicitly store memory content.
• evermemos_memory_profile
  • Get profile memory.
  • Update action currently returns TODO fallback until stable profile write schema is finalized.

CLI Commands

openclaw evermemos setup
openclaw evermemos status
openclaw evermemos search "<query>" --limit 6
openclaw evermemos wipe --yes

Slash Commands

/memory search <query>
/memory store <content>
/memory profile

EverMemOS API Used

• GET /health
• POST /api/v1/memories
• GET /api/v1/memories/search
• DELETE /api/v1/memories (for CLI wipe)

Development

npm install
npm run typecheck

One-Click Acceptance Script

Use the provided script on a real machine with OpenClaw + EverMemOS running:

./scripts/e2e-acceptance.sh

Optional environment overrides:

EVERMEMOS_BASE_URL=http://localhost:1995 \
EVERMEMOS_USER_ID=your-user-id \
OPENCLAW_CONFIG_PATH=/path/to/openclaw.yml \
REPORT_PATH=./TEST_REPORT_E2E.md \
./scripts/e2e-acceptance.sh

The script validates checklist items and outputs a Markdown report file.

Notes

• Local-first defaults are intentional (localhost:1995).
• Memory slot is exclusive in OpenClaw (kind: "memory").
• If EverMemOS is down, plugin degrades gracefully (no hard failure).