A persistent structured memory engine that enables AI agents to retain, deduplicate, and recall beliefs across sessions.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install openclaw-bee
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install openclaw-bee using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is openclaw-bee?
openclaw-bee, also known as the Belief Extraction Engine, is a powerful extension designed to provide agents with a long-term, structured memory. Unlike standard short-term context windows, this engine hooks into the lifecycle of your Openclaw Skills to extract meaningful data at the end of every session, ensuring that agents do not forget critical information between restarts.
By leveraging lightweight LLM calls for extraction and cosine similarity for deduplication, the engine maintains a clean and relevant database of beliefs. It allows for namespace isolation, meaning different agents like VECTOR or ORACLE can maintain their own distinct sets of knowledge, making it an essential tool for developers building complex, multi-agent systems.
openclaw-bee Use Cases
- Maintaining agent personalization and user preferences across multiple independent chat sessions.
- Creating specialized knowledge silos for different agent namespaces within a single environment.
- Reducing LLM token usage by injecting only the most relevant recalled beliefs rather than full histories.
- Tracking and managing subagent spawn budgets to optimize resource consumption in Openclaw Skills.
How openclaw-bee Works
- The engine initializes a persistent SQLite database to store agent beliefs and metadata.
- Upon agent termination, it triggers a background process to analyze the session transcript.
- A designated LLM (like Claude Haiku) extracts key 'beliefs' or facts from the conversation.
- New beliefs are compared against existing ones using cosine similarity to prevent redundancy.
- When a new agent session starts, the engine performs a semantic search to inject the most relevant beliefs into the system prompt.
openclaw-bee Setup
Install the extension globally via npm:
npm install -g @skysphere-labs/openclaw-bee
Configure your ~/.openclaw/openclaw.json to include the bee extension and define your database path:
{
"extensions": {
"entries": {
"bee": {
"enabled": true,
"config": {
"dbPath": "~/.openclaw/workspace/state/vector.db",
"agentId": "main"
}
}
}
}
}
Apply the changes by restarting the gateway:
openclaw gateway restart
openclaw-bee Data Schema & Taxonomy
The system organizes data within a SQLite database (vector.db) using the following structure:
| Component | Description |
|---|---|
beliefs table |
Stores the core text of extracted beliefs along with the confidence score and agentId. |
agentId |
A namespace string used to scope and isolate memory between different agents. |
vector_store |
Internal storage for semantic embeddings used in belief recall and deduplication. |
spawn_tracking |
Monitors the count of subagent spawns per session to enforce budgets. |
openclaw-bee Advanced Features
- Multi-agent scoping via unique agentId namespaces for isolated memory.
- Automatic semantic deduplication with a default 0.92 similarity threshold to merge similar facts.
- Configurable injection limits for core, active, and recalled beliefs to manage prompt space.
- Support for custom extraction models to optimize for cost or extraction accuracy.
- Integrated spawn budget warnings to prevent runaway agent loops.
SKILL.md
Loading
Related Openclaw Skills
Bluente File Translator
A professional-grade document translation skill for Openclaw Skills that automates the conversion of PDF, Word, and PowerPoint files via the Bluente API.
Deterministic Controller
A documentation-first template suite for building repeatable, evidence-gated orchestration loops in Openclaw Skills environments.
Agent Orchestrator
A sophisticated multi-agent orchestration framework that implements five proven patterns for coordinating specialized AI agents to solve complex problems.
Agent Orchestrator
A powerful multi-agent orchestration tool that implements five proven patterns to coordinate multiple AI agents for complex task resolution.
Fitness & Wellness Coach
A privacy-conscious AI coach that transforms fitness app screenshots into evidence-based training, nutrition, and recovery protocols.
ClawHub CLI
A dedicated command-line interface for searching, installing, and publishing agent capabilities within the Openclaw ecosystem.
Featured*
