I’m Evan, author of Harbor.
Have been working on this for about two years. Harbor started because I was tired of spending evenings debugging Docker compose files whenever I wanted to try a new local LLM project. It was supposed to be a weekend script, but the problem turned out to be bigger than I thought and it turned into a list of 129 services.
The basic idea is that local AI services should work together out of the box. To take shelter Gives you Open WebUI and Olama. You add more services to the name, like Harbor Up Searxng Speeches ComfuiAnd Harbor handles the wiring between them. SearXNG is already linked to your chat UI. Voice already works. Image Gen already works. No compose files to write, no ports to remember.
The catalog covers a lot of ground. It includes 16 chat and creative frontends (Open WebUI, LibreChat, SillyTavern, ComfyUI and others), 21 LLM backends (Olama, Llama.cpp, VLLM, SGlang, KTransformers and others) and web search, voice, RAG, vector databases, workflow engines like N8N and Defy, fine-tuning with Unsloth, Eval There are 90+ satellites for the harness. Monitoring, and much more.
Everything runs locally, talks to each other and is one To take shelter away. The latest release states port launchThat’s the feature I’m most excited about. It runs coding agents like Cloud Code, Codex, Copilot, and OpenCode against your local stack without touching a single NV version. Harbor Launch–Web Also gives them web search and opens up the entire pipeline behind the scenes. Harbor also exposes an OpenAI-compatible API, so any tool or SDK code that speaks OpenAI can point to your local model as a drop-in replacement.
MIT licensed, no telemetry, everything stays on your machine. CLI and desktop apps for Linux and macOS. If you run local models and are tired of manually linking things together, try this.
Happy to answer questions about architecture, supported services or how specific workflows work.
<a href