Agno
Agno is open-source software for building AI agents and running them as a live service. You can build single agents, teams of agents and step-by-step workflows.
Agno is open-source software for building AI agents and running them as a live service. An agent here means a program that gives a language model instructions, lets it ask for tools, and returns its answer. Agno also has a part called AgentOS. It runs those agents as a live service that people can use.
Picture an online shop. A customer asks when order ORD-123 will arrive. You write a normal Python function that looks up an order. Agno reads the function’s name, its short description and the kind of input it expects (text, a number and so on). It turns them into a tool definition, a short note the model can read. The model decides to call the tool with ORD-123. Agno checks the request, runs your function and hands the result back. The model then writes the reply.
One agent is not the only option. A team puts several agents under a leader. The leader can hand out tasks and combine the answers, or send a question to one specialist. A workflow runs agents through steps you define, such as one agent finding news stories and another writing an article. Memory lets an agent keep facts about each user, such as a preference for email, and recall them in a later chat.
AgentOS is the part that serves agents to real users. It offers an API (a way for other programs to send it requests), saved chat sessions, and links to chat apps such as Slack. Agno sends a usage report (telemetry) for each agent run. It leaves out prompts and outputs, and you can switch it off.
A working agent needs more than a model and a prompt.
Follow one delivery question through an Agno agent.
- 1 · describeAgno turns a Python function's name, docstring and type hints into a tool definition.
- 2 · sendThe agent passes its context, plus the list of tools, to the model.
- 3 · chooseThe model either answers or asks for one or more tool calls.
- 4 · runAgno checks the arguments, runs each tool and adds the results to the context.
- 5 · finishThis repeats, round after round, until the model gives its finished reply.
In Agno the model asks for a tool; Agno's code is what actually runs it.
| Who | What they ask | What it works with |
|---|---|---|
| Online shop | “When will my order arrive?” | An agent with an order lookup tool |
| Personal assistant app | “Can it remember that I prefer email updates?” | Memory stored per user in a database |
| Research team | “Can one agent find stories and another write the article?” | A workflow with defined steps |
| Support desk | “Can questions in different languages go to the right specialist?” | A team that routes to its members |
- It turns plain Python functions into tools a model can call.
- It can store facts about each user and bring them into later chats.
- It lets several agents work as a team or as workflow steps.
- Its AgentOS runtime serves agents through an API and chat apps.
- A team makes extra model calls, which adds waiting time and token use.
- A workflow's steps run in a set order, but agent outputs can still differ between runs.
- The docs advise using the same stable user ID for every run that should share memories.
- The docs advise a single agent when the task fits one area or token costs matter.
Sources used
This explainer is written in original language. The links below support its factual claims.
- repoagno-agi/agno, Agno (GitHub) · read 28 Sept 2026
- docsWhat are Agents?, Agno documentation · read 28 Sept 2026
- docsWhat are Tools?, Agno documentation · read 28 Sept 2026
- docsWhat are Teams?, Agno documentation · read 28 Sept 2026
- docsWhat are Workflows?, Agno documentation · read 28 Sept 2026
- docsWhat is Memory?, Agno documentation · read 28 Sept 2026
- docsWhat is AgentOS?, Agno documentation · read 28 Sept 2026