C/C++ header-only library for text embeddings, sparse embeddings, image embeddings, and document reranking via ONNX Runtime.

AI-native browser designed for the new era of computing. Secure browsing to protect privacy.

HuggingFace transformer compiler for optimised native inference binaries.

Token compression for LLM prompts via a learned token selector. Beats LLMLingua-2 by 4-5pp ROUGE-L.

Persistent memory layer for LLM agents. Zero API costs, sub-100ms ingestion, fully offline.

Synthetic LiDAR simulation platform for autonomous vehicle research.

Open-source alternative to Google Code Wiki — with MCP. Index codebases with tree-sitter, generate docs with LLMs.

Minimal tensor library with autograd for building deep learning models.

Stop overpaying for LLM calls. Automatically routes every prompt to the cheapest capable model.

Unified portable AI infra combining vectorisation, vector database, and LLM manager. Now open sourced.

Contributed to build the Zig library for building capable embedding indexings. High-performance vector search engine.