fish827-08/rag-kb

Overview

Local-first, offline-capable agent memory and knowledge service. Provides REST and MCP dual-protocol access with hybrid retrieval combining vector and BM25/RRF. Handles document and webpage RAG without requiring an LLM for storage and retrieval.

Best for

Best for
Developers needing a lightweight, offline knowledge service for agents

Use cases

• Build offline agent memory
• Serve hybrid retrieval endpoints
• Ingest documents and webpages for RAG

How to use

Install

pip install -r requirements.txt

Tools exposed

• KB_LLM_MODE
• KB_DEVICE
• KB_WATCH_DIR
• KB_DATA_DIR
• KB_EMBED_MODEL
• KB_LLM_MODEL
• KB_OLLAMA_BASE_URL
• KB_SENSITIVE_NAMESPACES
• KB_API_KEY
• KB_SPARSE_ENABLED

Tested with

Claude Code, Cursor, ChatGPT

Notes

Local-first, offline-capable agent memory and knowledge service. Provides REST and MCP dual-protocol access with hybrid retrieval combining vector and BM25/RRF. Handles document and webpage RAG without requiring an LLM for storage and retrieval.

2 stars on GitHub. Last updated 2026-09-07. Licensed Apache-2.0.

Use cases

• Build offline agent memory
• Serve hybrid retrieval endpoints
• Ingest documents and webpages for RAG

Pros

• Offline-capable, no LLM dependency
• Dual REST and MCP protocols
• Hybrid retrieval combines vector and BM25

Cons

• Low community adoption (2 stars)
• Limited documentation or support implied
• Python-only implementation

Indexed from awesome-mcp-servers-punkpeye and enriched against its public facts.