Loa

Loa is a local-first, stateful coding agent designed for project-wide engineering tasks and long-running autonomous work. It uses a mutable Directed Acyclic Graph (DAG) as an execution plan and context topology to enable complex, reliable macro-level architectural engineering. Loa operates with explicit state transitions, persistent task state, memory retrieval, evaluation, and recovery mechanisms and supports running locally or in a Docker sandbox with integration to OpenAI-compatible or local model inference endpoints.