council

Why we included this project

Building reliable multi-step AI workflows usually means routing requests across several specialized models instead of leaning on one prompt. Council, a Python framework, treats that orchestration as the core problem: you compose agents and govern how work moves between them, using controllers, evaluators, and filters to keep output predictable, plus budgets to limit what each step can spend. A unified interface sits in front of providers like OpenAI, Anthropic, Gemini, and local models served through Ollama, so swapping backends doesn't force a pipeline rewrite. Retries, error handling, and token and cost tracking are built in, which matters when you're wiring agents into a service that has to stay up. Teams already using LangChain or LlamaIndex can slot Council in alongside rather than rebuilding everything.