Show HN: LLMNet – The Offline Internet, Search the Web without the Web

Why It Matters

By keeping queries and data on‑device, LLMNet addresses growing concerns over data privacy and enables enterprises to maintain knowledge bases without internet dependency.

Key Takeaways

• Runs entirely offline, preserving privacy.
• Uses PostgreSQL pgvector with HNSW for fast search.
• Supports any local LLM via OpenAI‑compatible API.
• Indexes websites or wikis into persistent vector DB.
• Simple Bun‑based setup with Next.js UI.

Pulse Analysis

The rise of privacy‑centric AI tools reflects heightened awareness of data sovereignty, especially as enterprises grapple with regulatory pressures and the risk of data leaks. LLMNet positions itself at the intersection of offline capability and generative AI, offering a self‑contained search experience that eliminates reliance on external cloud services. This model appeals to organizations that must safeguard proprietary information while still benefiting from the rapid knowledge retrieval that large language models provide.

Technically, LLMNet combines a local Retrieval‑Augmented Generation (RAG) pipeline with PostgreSQL’s pgvector extension, employing HNSW indexing for sub‑second semantic queries. The ingestion workflow crawls target sites, converts content to clean Markdown, splits text via a recursive character splitter, and stores embeddings generated by a locally hosted LLM or embedding server. The front‑end, built with Next.js and Tailwind CSS, delivers a glass‑morphic UI, while Bun orchestrates dependency management and server startup, streamlining deployment for developers familiar with modern JavaScript ecosystems.

From a business perspective, the solution enables companies to create searchable internal knowledge bases without exposing data to third‑party APIs, reducing compliance risk and operational costs associated with cloud subscriptions. Its open‑source nature encourages customization, fostering adoption in sectors such as legal, healthcare, and finance where data confidentiality is paramount. However, performance hinges on the quality of the local LLM and hardware resources, suggesting that enterprises may need to invest in adequate compute infrastructure to fully realize LLMNet’s potential.