OpenClaw Memory Qdrant

Local semantic memory with Qdrant and Transformers.js. Store, search, and recall conversation context using vector embeddings (fully local, no API keys).

Security Scan

The plugin appears to do what it says: a local semantic-memory module using Transformers.js embeddings and optional Qdrant storage; its requirements and behavior are proportionate and documented.

Purpose & Capability

Name/description match the included code and metadata. Required binaries (node, npm), dependencies (@xenova/transformers, @qdrant/js-client-rest) and config options (qdrantUrl, autoCapture, maxMemorySize) are directly relevant to providing local embeddings and optional Qdrant storage. No unrelated credentials, binaries, or config paths are requested.

Instruction Scope

SKILL.md and the code focus on storing/searching memories, embedding text, and optional auto-capture. The docs explicitly disclose first-run model download (~25MB from Hugging Face) and native build requirements. The autoCapture feature can capture PII (emails/phone-like strings) and is documented as opt-in (default false) — this is a privacy risk if enabled but is stated. No instructions direct the agent to read unrelated secrets or system files.

Install Mechanism

There is no platform install spec in the registry (instruction-only), but package.json/package-lock and code files are included. Dependencies are pulled from npm (no suspicious external download URLs). The dependency tree includes native/optional native modules (sharp, onnxruntime, onnxruntime-web, undici) which may require build tools and increase installation friction. Model files are downloaded from Hugging Face at runtime (declared).

Credentials

The skill requests no environment variables or credentials. The only external connector is an optional qdrantUrl config — supplying a Qdrant server is expected for external storage. No unrelated SECRET/TOKEN env vars are requested. The plugin's documentation warns to only point qdrantUrl at trusted servers.

Persistence & Privilege

always is false and disable-model-invocation is false (normal). The plugin registers itself and may use lifecycle hooks for autoCapture/autoRecall, which is expected for a memory plugin. autoCapture is disabled by default. The skill does not request permanent system-wide privileges or modify other skills' configs.

Assessment

This plugin is internally consistent with its stated purpose, but consider the following before installing: - Test in an isolated environment first (it downloads a ~25MB model at first run and has native deps like sharp/onnxruntime that may need build tools). - Keep autoCapture disabled unless you trust the environment — enabling it can capture emails/phone-like strings and other PII. - If you configure qdrantUrl, only use a trusted Qdrant instance (data will be sent to that server). - Pin/install with npm ci or vendor dependencies for production to avoid unexpected upstream changes. - Review and monitor memory size settings (setting maxMemorySize to 999999 disables LRU deletion and can exhaust RAM in in-memory mode).

memory-qdrant

Use when you need your OpenClaw agent to remember and recall information across conversations using semantic search.

Local semantic memory plugin powered by Qdrant vector database and Transformers.js embeddings. Zero configuration, fully local, no API keys required.

Features

• Semantic search with local Transformers.js embeddings
• In-memory mode (zero config) or persistent Qdrant storage
• Optional auto-capture of conversation context (opt-in, disabled by default)
• Context-aware memory recall
• Fully local, no API keys or external services required

Installation

clawhub install memory-qdrant

First-time setup: This plugin downloads a 25MB embedding model from Hugging Face on first run and may require build tools for native dependencies (sharp, onnxruntime). See README for detailed installation requirements.

Configuration

Enable in your OpenClaw config:

{
  "plugins": {
    "memory-qdrant": {
      "enabled": true
    }
  }
}

Options:

• autoCapture (default: false) - Auto-record conversations. Note: trigger patterns include email/phone regex, so enabling this may capture PII.
• autoRecall (default: true) - Auto-inject relevant memories
• qdrantUrl (optional) - External Qdrant server (leave empty for in-memory)

Usage

Three tools available:

memory_store - Save information

memory_store({
  text: "User prefers Opus for complex tasks",
  category: "preference"
})

memory_search - Find relevant memories

memory_search({
  query: "workflow preferences",
  limit: 5
})

memory_forget - Delete memories

memory_forget({ memoryId: "uuid" })
// or
memory_forget({ query: "text to forget" })

Privacy & Security

• In-memory mode (default): Data cleared on restart
• Qdrant mode: Data sent to configured server (use trusted servers only)
• Network: Downloads ~25MB model from Hugging Face on first run
• autoCapture: Disabled by default for privacy. Trigger patterns match emails and phone-like numbers, so enabling autoCapture can capture PII.

Technical Details

• Vector DB: Qdrant (in-memory or external)
• Embeddings: Xenova/all-MiniLM-L6-v2 (local)
• Module: ES6 with factory function pattern

Links

• GitHub: https://github.com/zuiho-kai/openclaw-memory-qdrant
• Issues: https://github.com/zuiho-kai/openclaw-memory-qdrant/issues