Player not loading? Watch on YouTube
This GitHub project roundup covers local AI apps and development utilities alongside games, interface components and other software. It introduces projects rather than walking through installation or testing them against each other.
The speaker describes ReScript as a browser editor that cuts audio or video when you delete transcript words. Whisper generates timestamps locally; media stays on the device, but transcription requires a large initial model download, and Chromium has the best support. Lexicon combines LanguageTool grammar checks with optional on-device models for rewriting and summarizing. For macOS, Yap uses speech APIs introduced in macOS 26, while Quill records microphone and system audio separately and transcribes both locally. The speaker notes that Quill has no license file.
Agent utilities include agentacct, which reads Claude Code and Codex logs locally, joins usage with recorded tasks and test evidence, and separates verified work from agent reports without provider keys, and OptMem, which stores persistent memory in append-only text files. open-science keeps projects on the computer, but sends prompts and selected context to configured model providers.
DeltaFin is the roundup's unusual attempt to run models locally. The speaker reports Kimi K3 running on a 64 GB MacBook by fetching experts from Hugging Face into a cache. At about one token per minute, the author describes it as an existence proof rather than a practical chat setup. Its network dependency also separates it from an offline model runner.
