A Karpathy-style LLM wiki your agents maintain (Markdown and Git)
Markdown + git beats pgvector for agent memory — refreshingly simple architecture.
Chat with the just released ufo files. Vector RAG setup with pre-built embeddings database using sqlite. Ready to chat.
Yet another chat-with-docs tool, just with UFO files pre-loaded instead of your own.
UFO enthusiasts, developers testing RAG setups
ChatPDF · DocuChat · PrivateGPT
Contains sqlite database with all files pre-loaded along with embeddings. Has good looking UI you can directly chat with.
Demo(for just next 15 minutes or I run out of 5$ credits). Also, concurrency is limited by openrouter to just 10... so, some requests might fail:
https://open-proxy.space
^ Im running this on my machine begind cloudflare proxy.. and this is the domain I generally use for all dev purposes..
Markdown + git beats pgvector for agent memory — refreshingly simple architecture.
Useful calculator, but spreadsheets and existing tools already do this.
Using a single-file .pardus format with CREATE/INSERT/SELECT + SIMILARITY queries gives a very familiar developer UX for embedding storage. The combination of graph-based ANN, full transactions, thread-safety, and zero external dependencies is an uncommon and useful engineering combo for local-first AI work; it would win more attention with benchmark comparisons and richer ecosystem integrations (connectors/clients).
Breaks down hidden RAG costs like vector storage overhead and HNSW indexing fees.
Columnar storage inside SQLite delivers 130,000x speedup on aggregation scans.
Branchless CUDA guardrails sound fast, but semantic safety needs context, not just bitwise math.