deeplake-memory

Cloud-backed agent memory plugin for OpenClaw powered by DeepLake.

Install

# Install DeepLake CLI and set up your mount
curl -fsSL https://deeplake.ai/install.sh | bash
deeplake init

# Install the OpenClaw plugin
openclaw plugins install deeplake-memory

Then enable the write tool:

// In openclaw.json
{ "tools": { "alsoAllow": ["memory_store"] } }

Restart the gateway to apply.

How it works

DeepLake provides a cloud-backed FUSE filesystem. Files written to the mount sync to DeepLake's cloud in real-time and persist across sessions, machines, and agents.

The plugin provides three tools:

• memory_search grep-based search over memory files
• memory_get read a specific memory file by path
• memory_store write to a memory file

Plus automatic hooks:

• Auto-recall searches and injects relevant memories before each agent turn
• Auto-capture saves conversation context before compaction

The plugin auto-detects your DeepLake mount from ~/.deeplake/mounts.json. No API key or additional config needed.

Configuration

All config is optional the plugin works with zero config if DeepLake CLI is installed.

// In openclaw.json → plugins.entries.deeplake-memory.config
{
  "mountPath": "/path/to/mount",  // Override auto-detected mount path
  "autoCapture": true,            // Auto-save memories before compaction
  "autoRecall": true              // Auto-inject memories before each turn
}

Why DeepLake over LanceDB/OpenViking?

• Cloud-native no local server, no local database. Your data lives in DeepLake's cloud.
• Zero deps no WASM, no S3 SDK, no embedding API keys. Just a filesystem.
• Works offline FUSE mount caches locally. Syncs when connected.
• Works with local models no API key needed. Qwen, Llama, etc. all work.
• Agent-native agents already know how to read/write files. No custom tooling required.

License

MIT