memorylake-openclaw plugin connects OpenClaw (and QClaw) agents to MemoryLake. This page is the configuration and capability reference; for a narrative walkthrough see the Coding Agents scenario.
Installation
- One-line installer (recommended)
- Manual
The console’s Integrations → OpenClaw card generates this pre-filled:Windows PowerShell: use
install.ps1 from the same host. The script writes the plugin config, sets tools.profile to "full" (required; it asks first if you’ve customized it), and restarts the gateway. Environment variables MEMORYLAKE_API_KEY / MEMORYLAKE_PROJECT_ID / MEMORYLAKE_HOST can replace the flags.Configuration Reference
Per-agent / per-directory override
{workspace}/.memorylake/config.json overrides the global config for sessions in that directory — typically a different projectId per client or codebase. The bundled agent-memorylake-config skill sets this up conversationally.
Automatic Behaviors
- Auto-Recall — injects instructions (plus a per-turn reminder) directing the agent to call
retrieve_contextbefore answering each user message, and announces the project’s subscribed Open Data categories at session start. Model-driven recall keeps latency low while ensuring fresh context. - Auto-Capture — after each successful user turn, new conversation messages (deduplicated via a per-session watermark) are sent to MemoryLake, where extraction decides what to store, update, or merge. Tool outputs are not captured.
- Auto-Upload — files you share in chat are detected and uploaded to the project asynchronously (deduplicated against prior uploads); the agent is never blocked waiting.
Agent Tools
CLI
upload accepts single files, directories (500-file guardrail against accidental node_modules uploads), and archives — zip/tar/tgz/7z/bz2/rar/xz are auto-extracted and uploaded with concurrency, skipping .DS_Store-style junk.
Bundled Skills
Companion: Models Plugin
memorylake-models-openclaw (separate plugin) registers MemoryLake Model Router as an OpenClaw model provider — one sk-… key for all supported LLMs:
GET /v1/models; the plugin writes an OpenAI-compatible provider entry (https://app.memorylake.ai/v1) into your OpenClaw config.
Privacy & Telemetry
The plugin talks only to your configuredhost. Usage telemetry is opt-in — nothing is reported unless you explicitly configure a telemetry key.
Troubleshooting
- Tools missing → ensure
tools.profileis"full"and the gateway was restarted after install. - 401 errors → key invalid/expired; update
apiKey(hot-reloads, no restart needed). - Wrong project receiving memories → check for a per-directory
.memorylake/config.jsonoverriding your globalprojectId.