slowburn 0.3.0

Cost-Sustainable Concurrent Execution for Long-Horizon LLM Agents

SlowBurn

SlowBurn: Cost-Sustainable Concurrent Execution for Long-Horizon LLM Agents

SlowBurn is a Python library that lets LLM agent workflows run within a dollar budget by automatically slowing down — not crashing — when the budget is tight. It combines concurrent execution with dollar-denominated rate limiting and per-call cost tracking, so researchers can launch overnight batch experiments on a $5/day budget and wake up to completed results instead of a crashed script.

SlowBurn provides a native asyncio LLM worker for direct use, and drop-in integrations for AutoGen (AG2) and CrewAI that add budget control to existing multi-agent workflows without code changes.

Installation

pip install slowburn

Development

git clone https://github.com/adivekar-utexas/slowburn.git
cd slowburn
pip install -e ".[dev]"
pytest

License

MIT