agent-framework

Why we included this project

Once a team outgrows single-prompt demos and starts putting agents into real production flows, this framework earns its keep. It gives Python and .NET teams the same agent abstractions, so a C# backend and a Python service can share orchestration logic instead of each side reinventing its own. The workflow engine handles the patterns that actually come up in production, including sequential pipelines, concurrent fan-out, handoffs between specialized agents, and group collaboration, and it adds checkpointing and human-in-the-loop control so long-running work can pause, restart, or get reviewed. Observability and versioning are covered through OpenTelemetry tracing and YAML-defined agents, and because the design is provider-agnostic, you can start with one LLM vendor and switch later without touching your orchestration layer. For multi-agent systems with durability and governance requirements, this is a solid foundation to build on.