mok is a terminal-based coding agent harness built in Go. It talks to OpenAI-compatible LLM endpoints with SSE streaming and handles tool calls (read, write, edit, bash), thinking, and loops.

With a strong focus on running perfectly with non-frontier and local models, mok includes quirk handling for tool calls, empty responses, loops, JSON repair...

Its multi-agent orchestration lets you define agents with their own prompt, model, and config overrides, then chain them into flows that run sequentially with handoff summaries between isolated contexts.

Additional features include automatic context compaction with LLM-driven summarization, session save/restore, slash commands for model switching and YOLO mode, and configurable allowlist/blocklist for bash command execution.