ChronOllama
An experimental autonomous software-building agent running entirely on local models served through Ollama. A conversational phase clarifies the user's request, a planner model breaks it into dependency-aware steps, and worker models generate the corresponding files in parallel where possible. Every file passes through a linting retry loop, a testability check, and either automated test generation/execution or a fallback model evaluation, with failures triggering retries and, past a threshold, replanning around the failed step. The system swaps between detailed and simplified prompt sets depending on the parameter size of the local model in use.