A high-performance local knowledge base skill utilizing FAISS and BM25 for hybrid search powered by Alibaba Cloud Bailian embeddings.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install rag-kb
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 rag-kb 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 Bailian FAISS Knowledge Base?
The Bailian FAISS Knowledge Base is a sophisticated tool designed for the Openclaw Skills ecosystem to maintain file-based local knowledge repositories. It bridges the gap between raw text extraction and intelligent retrieval by combining FAISS for dense vector search and BM25 for sparse keyword search. This hybrid approach ensures that AI agents can find relevant information with high precision, whether through semantic meaning or specific terminology.
By leveraging Alibaba Cloud's text-embedding-v4 and optional qwen3-rerank models, this skill provides industrial-grade RAG (Retrieval-Augmented Generation) capabilities. It is built to work seamlessly with the Openclaw file processing workflow, consuming extracted chunks and Text-to-Question (T2Q) data to build a robust index that resides entirely on your local infrastructure while only calling external APIs for embedding and reranking tasks.
Bailian FAISS Knowledge Base Use Cases
- Building a local searchable repository from thousands of technical documents or regulations.
- Implementing hybrid search (semantic + keyword) to improve retrieval accuracy for AI coding assistants.
- Managing document versions using timestamped directory structures within Openclaw Skills.
- Adding protected terms to local indices to ensure specialized industry jargon is correctly indexed for keyword search.
- Enabling high-precision RAG by reranking initial search results with advanced LLM-based rerankers.
How Bailian FAISS Knowledge Base Works
- OpenClaw extracts text from source files and organizes them into chunks, summaries, and T2Q (Text-to-Question) files.
- The Python script processes these files, generating 1024-dimensional vectors using the Alibaba Cloud Bailian text-embedding-v4 model.
- A semantic index is generated via FAISS and stored locally as index.faiss, while a keyword index is created using BM25 and stored as bm25.json.
- When a query is initiated, the skill performs a hybrid search across the specified knowledge base or across all available repositories.
- Results from semantic and keyword searches are fused, and if enabled, a reranking step is performed to prioritize the most relevant context for the AI agent.
Bailian FAISS Knowledge Base Setup
To get started with this skill within the Openclaw Skills framework, follow these steps:
# Install the required Python dependencies
python3 -m pip install -r requirements.txt
# Set your Alibaba Cloud Bailian API Key
export BAILIAN_SK="your_api_key_here"
# Verify your environment configuration
python3 scripts/bailian_faiss_kb.py doctor --root-dir /var/openclaw-kb
Bailian FAISS Knowledge Base Data Schema & Taxonomy
The skill follows a strict directory and metadata taxonomy to ensure reliable indexing and retrieval:
| File/Folder | Purpose | Format |
|---|---|---|
summary.txt |
Single-line summary of the document for quick context. | Plain Text |
chunks/*.md |
Semantic text segments used for primary indexing. | Markdown |
t2q/*.md |
Generated questions representing the chunk for better recall. | Markdown |
vectors.jsonl |
Local cache of generated embeddings and metadata. | JSONL |
index.faiss |
The binary vector index for FAISS. | Binary |
bm25.json |
Inverted index used for keyword-based retrieval. | JSON |
protected_terms.json |
Custom vocabulary to protect specific terms during tokenization. | JSON |
Bailian FAISS Knowledge Base Advanced Features
- Hybrid retrieval mode combining semantic FAISS search and BM25 keyword matching for maximum recall.
- T2Q (Text-to-Question) indexing where questions serve as semantic proxies for content chunks.
- Support for qwen3-rerank to provide high-precision document ranking.
- Local management of protected terms to refine the jieba segmentation process.
- Global query support to search across multiple independent knowledge bases simultaneously.
- Robust safety boundaries with no background services or listening ports, using only secure HTTPS requests to official API endpoints.
SKILL.md
Loading
Related Openclaw Skills
X Content Mentor
A strategic advisor that transforms raw ideas into high-performing X-native content while diagnosing growth bottlenecks for creators and indie builders.
A-Share Paper Trading Skill
A high-fidelity simulation and backtesting engine for the A-share market that strictly enforces T+1 settlement and price limit rules.
Paul Graham Perspective
An advisor-first skill that applies Paul Graham's core mental models—like writing as thinking and founder-over-idea judgment—to complex decision-making.
Naval Perspective
A strategic advisor skill that applies Naval Ravikant's core mental models to career, wealth, and decision-making challenges.
eNSP Topology Generator
A specialized tool for generating native Huawei eNSP topology files for network simulation and design.
AI Podcast Dual Host Generator
A professional AI-driven pipeline that transforms any text, URL, or PDF into a structured two-person dialogue podcast with stereo TTS audio.
METADATA
Featured*
