Skip to content

Python Agent Runtime

python-agent-runtime is a small execution engine for AI agents. It coordinates instructions, history, optional memory, optional retrieval, tools, callbacks and an LLM adapter.

The project is intentionally limited. It is meant to be a clean portfolio-grade Python library, not an orchestration platform or another full agent framework.

What It Provides

  • A compact Agent facade.
  • A deterministic Runtime execution loop.
  • Prompt construction from composable inputs.
  • Conversation history and optional memory.
  • Tool registration and execution.
  • Lifecycle callbacks.
  • Streaming responses.
  • Optional integration with python-rag-framework.
  • Callable and OpenAI LLM adapters.

Relationship to python-rag-framework

python-rag-framework owns retrieval: ingestion, chunking, embeddings, vector stores, reranking and citations.

python-agent-runtime owns agent execution. When a rag.RAG instance is provided, the runtime calls RAG.retrieve(...) and inserts retrieved chunks into the agent prompt. It does not reimplement retrieval or call RAG.ask().

Non Goals

This project does not implement multi-agent systems, workflow graphs, DAGs, schedulers, distributed execution, servers, dashboards or cloud deployment.