Install
pip install -U agno openai# serving: pip install -U "agno[os]"Agno: agents, teams and workflows in Python, served with AgentOS.
Agno is a Python framework for building agents, teams of agents and step-based workflows, together with AgentOS, a FastAPI runtime that serves them over REST, SSE and MCP with storage, authentication and per-user isolation. Agents, teams and workflows share one run interface and one storage layer, so moving from a script to a service is mostly a matter of handing them to AgentOS.
Install
pip install -U agno openai# serving: pip install -U "agno[os]"Version
agno 3.2.0 · Python 3.9+ (package metadata) Pages are checked against this release (2026-11-19).
Best for
Python services that need agents with sessions, memory and knowledge, multi-agent teams, and an HTTP API you run yourself.
Agno fits when you want one Python library to cover the agent, its storage and the service around it. It has built-in sessions, user memory, knowledge bases, guardrails, human-in-the-loop pauses, multi-agent teams and deterministic workflows, and AgentOS adds a ready-made API, MCP server and chat interfaces on infrastructure you control.
Consider something else when:
Agno’s surface is large and changes quickly (ten patch releases in the three weeks before this check), so pin the version and read the release notes before upgrading.
from agno.agent import Agentfrom agno.models.openai import OpenAIResponses
def get_weather(city: str) -> str: """Return the current weather for a city.""" return f"Sunny in {city}"
agent = Agent( model=OpenAIResponses(id="gpt-6.1-sol"), tools=[get_weather], instructions="Answer in one sentence.",)run = agent.run("What's the weather in Leeds?")print(run.content)Serve the same agent as an API:
from agno.agent import Agentfrom agno.db.sqlite import SqliteDbfrom agno.models.openai import OpenAIResponsesfrom agno.os import AgentOS
agent = Agent(id="assistant", model=OpenAIResponses(id="gpt-6.1-sol"))agent_os = AgentOS(agents=[agent], db=SqliteDb(db_file="agno.db"))app = agent_os.get_app() # uvicorn app:app, then POST /agents/assistant/runsReady? Start Zero → Hero step 1 →
Condensed release history; dates are first PyPI uploads. Agno was previously published as phidata. The upstream v3 changelog and v2 changelog have the detail.
| Version | Date | Highlights |
|---|---|---|
| 1.0.0 | 30 Jan 2025 | First 1.x release under the Agno name (a 0.1.0 had been uploaded on 23 Jan). |
| 1.1–1.8 | Feb–Aug 2025 | Minor releases in the 1.x line; 1.8.x patches continued to 25 Sep 2025. |
| 2.0.0 | Sep 2025 | Major version. agno.storage classes became agno.db classes, one instance shared by agents, teams and workflows; Playground replaced by AgentOS; Apps renamed Interfaces; Workflows rebuilt around steps; Memory, Knowledge, Teams and metrics reworked. |
| 2.1–2.9 | Oct 2025 – Aug 2026 | Minor releases in the 2.x line (2.9.0 on 13 Aug 2026). |
| 3.0.0 | Aug 2026 | Major version. Runs stored one row per run (agno_runs) with MigrationManager; user_id isolation across memories, knowledge, evals, metrics and schedules; durable background queue (QueueConfig); workflow HITL moved to HumanReview; reasoning=True, MultiMCPTools, culture and several deprecated parameters removed. See Upgrading from v2. |
| 3.1.0 | Oct 2026 | Minor release, followed by patch releases up to 3.2.0. |
| Date | Summary |
|---|---|
| 2026-10-24 | Guide written for agno 3.2.0: overview, comprehensive guide, teams, workflows, AgentOS, recipes, production and diagrams. Examples run offline with a scripted model. |
| 2026-10-28 | Updated to agno 3.2.0. Added Skills, user-control-flow and user-feedback tools, chunking strategies and the recency and MMR rerankers to the comprehensive guide. Scripts and the offline knowledge examples were run; no vector database was available, so Knowledge.insert(reader=...) with a store was not run. |