Local-first AI that swims with you. Bring your own models, route work between agents, and keep every byte in waters you control.
Three principles underneath. The rest of the system flows from those.
Engineers' laptops, idle workstations, the old Mac Studios in the lab. Shoal puts them on shared duty without buying a single new accelerator.
Models run on your hardware. Inference, embeddings, prompts, conversation history — none of it leaves the boundary you set, ever.
OpenAI- and Anthropic-compatible APIs on the front. Plain Ollama, vLLM, llama.cpp, MLX on the back. Glue between parts you already trust.
Three tiers, one protocol, your hardware. Add or pull a worker — the rest of the system never notices.
One container brings up the server and dashboard. Point a worker at it from anywhere on your fleet.
$ docker pull ghcr.io/nitrictech/shoal-lite
$ docker run -p 3000:3000 ghcr.io/nitrictech/shoal-lite