Agno is a Python framework for building agents that ships with a runtime and a control plane. It covers everything from a single agent to coordinated agent teams and stateful workflows.
What it is for
- Building agents with tools, memory and a knowledge base
- Assembling agent teams with roles and coordination between them
- Publishing agents as an API or MCP server for other systems to consume
- Running agents in production with sessions, traces, evaluations and approvals
How it works
- The agent is defined in Python with a model, instructions, tools and memory
- AgentOS turns the code into a service with an API, scheduler and dashboard
- Sessions and memory persist history and preferences across runs
- Flows allow deterministic steps combined with agent steps
- Control layers cover cost, limits, roles and an audit trail
Availability
- Open-source library under the Apache-2.0 licence (previously Phidata under MPL-2.0)
- The project was renamed from Phidata to Agno; the
phinamespace was replaced byagno
Points to note
- Existing Phidata projects must migrate package names and imports
- The runtime and dashboard pay off more in production than in prototypes
- Compare with more minimalist frameworks if you only need simple orchestration