mini-swe-agent - Minimal ~100-line SWE agent that solves GitHub issues or assists in the terminal; built for experimenting with agent design.

Jina Reader - URL-to-clean-content reader (r.jina.ai) that returns simplified, LLM-ready page text from any web page.

SearXNG - Self-hosted, privacy-respecting metasearch engine that makes a handy retrieval backend for AI agents.

MLX-VLM - Local inference and fine-tuning of vision-language models on Apple Silicon via MLX, with speculative decoding and prefix caching.

Bumblebee - Perplexity's open-source read-only scanner for macOS/Linux dev machines that inventories risky packages, extensions, and tool configs.

SmithDB - LangChain's data layer for agent observability: large nested traces on object storage, built on Rust, DataFusion, and Vortex.

Repo2RLEnv - Turns repos, PRs, commits, and CVEs into runnable, verifiable coding environments with programmatic rewards for agent eval or RL.