ComfyUI, ds4 and OmniRoute: local AI tools overview
Explore local inference, coding agents and generation tools, including ds4's DeepSeekV4 focus and ComfyUI workflows on Windows, Linux and macOS.
Player not loading? Watch on YouTube
This roundup introduces 20 developer projects, with sections on local inference, coding agents and visual generation alongside general development utilities. It explains their intended uses rather than walking through installations or comparing benchmark results.
The presenter describes ds4, also called Dwarf Star, as a dedicated engine for DeepSeekV4 Flash and Pro rather than a general local LLM runner. Its discussion covers high unified memory hardware, SSD-backed KV state and inference split across machines. OmniRoute takes a different role: a self-hosted gateway that routes requests to model providers through an OpenAI-compatible endpoint. Running that gateway locally does not mean the models run on the same machine.
JCode's section covers semantic memory, parallel agents and connections to Ollama or LM Studio. Code Review Graph and Code-Graph-RAG organize repository structure for a coding assistant, using SQLite and Memgraph respectively. Pi Web provides a browser workspace for Pi sessions; the presenter notes that Pi itself has no built-in sandbox by default.
ComfyUI's section explains reusable node workflows and local operation on Windows, Linux and macOS, with optional API nodes for closed models and a separate hosted cloud option. OpenEdit has a narrower requirement: an Apple silicon Mac. Its transcription uses a Veed account and workspace credits, while the remaining editing work runs locally.