
Ollama is an open-source program that downloads and runs open AI models on your own computer. It serves them on a local address that your code can call, with no API key for local models.
Official libraries listed in the Ollama repo: ollama-python and ollama-js.
Ollama v0.40.16: /v1/responses follows truncation and no longer cuts input silently, and /v1/models returns [] instead of null.
Ollama v0.40.14: OpenAI-compatible calls with no temperature or top_p now use the model's own values, not 1.0.
Ollama v0.40.15 runs MiMo V2.6, a 9B code and agent model, and AutoTrust models that return yes/no probabilities and choices.
Flint runs on Linux and Windows from v0.40.9. v0.40.11 adds Phonon speech on macOS via /v1/audio/speech and d3 decision models.
A 120B vision-language model (12B active, 1M context) and a 74B Bolt version that fits a 24 GB NVIDIA GPU, local in Ollama.