Local AI, without the setup
dotcore loads models, fits them to your hardware, and puts a single endpoint in front of everything.
Pick a model, get a running endpoint
Choose a model, or pull one straight from Hugging Face, and dotcore starts it for you: a build that matches your GPU, a sensible quantization, up behind an OpenAI-compatible endpoint. No engine flags, no driver versions to match by hand.
Hugging Face
Llama
Mistral
DeepSeek
Qwen
GemmaFit the model to your hardware
dotcore looks at the hardware you have, the context length you want, and the way requests arrive, then tunes the model and the engine to your workload. You get the most out of the machine you already paid for, without turning knobs.
Built on the engines you already know
dotcore runs on top of engines like vLLM, llama.cpp, MLX, and Ollama. It isn't a new engine: it picks the right one for each model and machine, sets it up, and keeps it running, so your models run on the projects the community already trusts.
vLLM
llama.cpp
Ollama
OllamaOne endpoint in front of everything
Every model and every machine sits behind a single endpoint. Requests go where they should, and your apps keep pointing at the same address even as you start, stop, and move models around. It can also reach past your own machines to cloud providers like OpenAI and Anthropic, so one address covers models running at home and in the cloud.
Many machines, working as one
Begin on a single laptop, then add a second, or a few more, at home or in the cloud. They don't have to sit in the same room: machines in different places, and even a friend's, can join the same pool of compute behind a single endpoint.
An app catalog, too
Open-source chat UIs, RAG tools, coding agents, and more that install and connect to your models in one step, all running on your own hardware.
Open WebUI
LibreChat
AnythingLLM
OpenClaw
Hermes Agent
Claude Code
Cline
opencode
PI