原文
The Offline Internet. A premium, private, and AI-powered search experience that lives entirely on your machine.
LLMNet transforms your local LLMs into a structured search engine. It combines the power of local generative AI with a high-performance Vector Database (RAG) to provide instant, offline answers from your own knowledge base.
Home Page - Screenshot
- 🔒 100% Private: Your queries and data never leave your local network.
- 🧠 Local RAG: Index any website or wiki into a persistent Postgres Vector DB.
- ⚡ Instant Results: Sub-second semantic search using pgvector & HNSW indexing.
- 🎨 Premium UI: A glassmorphic, dark-mode interface inspired by modern search engines.
- 🌐 No Internet Required: Once indexed, your knowledge stays available offline.
- Frontend: Next.js, Tailwind CSS
- Intelligence: Local LLMs (via OpenAI-compatible APIs)
- Database: PostgreSQL with
pgvector - Orchestration: Bun, Cheerio (Crawl), Turndown (Markdown)
Ensure you have the following running locally:
- LLM Server: Port configured in
.env(e.g., Llama.cpp, Ollama) - Embedding Server: Port configured in
.env - Database: Postgres with the
vectorextension (seepostgres-pgvector/)
Configure your environment variables in .env:
# Example configuration
API_BASE_URL=http://localhost:8888/v1
EMBEDDING_URL=http://localhost:8889/v1/embeddings# Install dependencies
bun install
# Initialize Database
bun postgres-pgvector/migrate.ts
# Start the engine
bun devVisit localhost:3000 to start searching.
LLMNet features a recursive ingestion pipeline. Simply paste a documentation URL or a GitHub Wiki link into the Indexer, and the system will:
- Crawl the site (Recursive BFS).
- Convert content to clean Markdown.
- Chunk text using a Recursive Character Splitter.
- Embed & Store vectors for semantic retrieval.
Built for those who value privacy and data sovereignty.