Loom

Loom (loom-threads) is a thread-and-episode orchestration agent harness on its own minimal agent engine over any-llm providers. An orchestrator LLM plans and decomposes work but is deliberately tool-restricted: it can only dispatch threads. Each thread runs one bounded action in a fresh worker context (file/bash tools, jailed to a working directory) and returns an episode — a compressed record of what was done. Episodes are the only data that crosses between threads; nothing else leaks into orchestrator or worker context.

uv tool install loom-threads
loom setup  # provider config (once)
loom "add rate limiting to the API"

From source:

uv sync
uv run loom "add rate limiting to the API"

Docs: wiki/quickstart.md