Player not loading? Watch on YouTube
GitHub Trending Weekly #40 surveys 35 projects, with several focused on local AI and coding workflows. The speaker describes clawk as a disposable Linux VM for coding agents: it mounts the repository while keeping the host filesystem and keychain out of reach. clodex-ide takes a different approach, with persistent tasks, isolated agent and MCP processes, and approval gates for high-impact actions.
Memory tools receive substantial attention. paxm starts with local SQLite without keys or embeddings and can later use Zep or Mem0. understory stores agent knowledge as linked Markdown concepts; the speaker says it works with local models. deja-vu indexes existing coding session logs, while personal-model reads macOS accessibility data and uses on-device OCR as a fallback. The presentation describes inspectable memory, encrypted screenshots and no telemetry for personal-model.
Waku is presented as a local-first personal AI agent with a roughly 95-line Python core loop, SQLite memory and a dashboard for tool calls and costs. Inference School takes an educational route through Swift and Metal kernels on Apple Silicon, with 47 runnable problems and a 545-page companion book. neverclick uses local computer vision for keyboard navigation across Windows apps and is described as entirely offline.
These are brief project introductions rather than installation walkthroughs or independent benchmarks. Local storage and execution features do not establish that every project's model inference runs on the user's hardware.
