AgentJIT

AgentJIT is an open-source Python library that compiles dynamic, multi-step large language model (LLM) agent workflows into highly optimized, deterministic Python code. It achieves up to 100,000x speedup and eliminates token costs by running compiled code locally. It transparently bails out to the dynamic LLM agent if input diverges, ensures multi-core safe execution, and integrates with AI frameworks like LangChain, CrewAI, and OpenAI tools.