Spec Kit, RAGFlow and Unsloth: 20 developer tools

This roundup covers open source developer projects, with several aimed at people who run models locally or manage coding agents on their own machines. It describes capabilities and installation options rather than presenting comparative benchmarks.

The speaker presents Unsloth as a tool for model inference and fine-tuning, with a desktop app, Studio web interface and Python core. The overview includes GGUF export and compatible APIs for connecting coding agents. Claimed training speed and memory benefits are not measured in the transcript.

oMLX receives a more platform-specific introduction: an inference server for Apple Silicon, managed through the macOS menu bar. The speaker lists M1 through M4 support and describes a RAM-and-SSD cache that survives server restarts. llmfit addresses hardware selection by detecting CPU, RAM and GPU resources, recommending quantization and estimating generation speed.

For a self-hosted AI agent workflow, the roundup describes RAGFlow's document retrieval and citations, OpenViking's persistent context store, and Paritok's context compression proxy. The speaker says Paritok can run on a single 24 GB GPU or use a hosted endpoint. Spec Kit structures development around specifications, while T3 Code controls agents already installed and authenticated on the user's computer. Other segments cover security skills, interface rules and broader developer utilities.