Agently
A Python runtime framework for building GenAI application backends with structured outputs, observable actions, reusable skills and Model Context Protocol support.
Agently is a runtime framework for building AI application backends that need to behave reliably in production rather than just working once in a demo. It defines Actions and Skills that preserve typed execution, giving an application stable contracts, observable execution and restart-safe workflow boundaries around model calls, so behavior stays useful even as the underlying model changes.
It targets product engineers building assistants, internal copilots and AI-backed APIs, and platform teams that need extension points for model providers, tools, MCP servers, sandboxes, workflows and observability. Recent releases add retry-stream replay, fail-closed model aliases, execution-scoped skills, and programmatic Action calling, with typing for agent execution appearing directly in an editor. It supports many model providers, including OpenAI-, Anthropic- and Google-compatible APIs as well as Chinese providers such as ChatGLM, ERNIE and MiniMax, and is licensed under Apache 2.0.