ollaya
A local runtime, modeled on Ollama, for pulling and serving small decision models that turn a piece of text plus typed questions into calibrated probabilities in milliseconds, through a TypeSafe-compatible API.
ollaya runs open decision models locally the way Ollama runs LLMs. A decision model takes a state (a message, email or ticket, expressed as JSON) and typed questions such as choice or score, and returns calibrated probabilities in a single forward pass without generating text. The ollaya binary provides serve, run, pull, list, ps, show, rm, cp, stop and create commands, starting its own daemon on demand.
It speaks TypeSafe's /v1/systemone wire format as well as a native API, so existing TypeSafe clients work by changing one base-URL variable. Model weights are not re-hosted: ollaya publishes small ONNX graphs that read the original weights from the author's Hugging Face repository, pinned by commit and verified by sha256, or run the author's own GGUF file through llama.cpp. It also serves models to Claude Code, Claude Desktop, Cursor and other MCP clients.