# Agno — An Apache 2.0 Python agent SDK and AgentOS runtime, with a Control Plane you run in your own cloud.

> Source: The Agents Index — https://theagentsindex.com/agno (structured, researched, re-verified)
> Facts last verified: 2026-08-23

Agno is an open-source Python SDK for building agents, teams and workflows, paired with AgentOS, the runtime that serves them, and the AgentOS Control Plane, a web UI you connect to from your browser. The SDK and AgentOS are Apache 2.0 and free. The paid Pro tier adds a live hosted Control Plane at a real URL with shared, role-scoped access. Enterprise adds a self-hosted Control Plane, SLAs and a dedicated technical lead. Founded in 2023 as Phidata, the project rebranded to Agno in January 2025 and ships as Agno 3.0 today. The codebase sits under the agno-agi organisation at 41,839 GitHub stars and 5,807 forks.

| Fact | Value |
| --- | --- |
| Website | https://www.agno.com |
| Pricing | Free SDK and AgentOS, Apache 2.0. Free Control Plane runs locally. Hosted Pro Control Plane $150/month with shared, role-scoped access. Enterprise is custom and adds a self-hosted Control Plane option. |
| API | Yes |
| Best for | Python developers who want an Apache 2.0 SDK with memory, knowledge, learning, guardrails and a Control Plane they run inside their own cloud account, and who are comfortable running Docker or a cloud deploy template themselves. |
| Not for | Teams that want a fully managed, zero-infrastructure cloud runtime with no cloud account of their own to operate. CrewAI's managed cloud or a no-code agent platform fits that need better than Agno's paid tier, which still requires running your own Postgres and container host. |

## Pricing

| Tier | Price |
| --- | --- |
| Free | Free |
| Pro | $150/mo |
| Enterprise | Custom (contact sales) |

## Verdict

Agno is a fast-moving open-source project that started in 2023 as Phidata, a data-engineering toolkit, became Agno in January 2025 and now ships as three pieces: the free Apache 2.0 Agent SDK, the free AgentOS runtime, and the AgentOS Control Plane, which is local and free on the developer machine and becomes a live hosted URL on the paid Pro tier. The current version (Agno 3.0) is positioned as "built for coding agents" and ships its docs as an MCP server at docs.agno.com/mcp so a coding agent can scaffold a platform from documented parts. The SDK page quotes a 3.2-microsecond agent instantiation and 5.2 KiB of memory per agent, measured on a public harness the team publishes, which is more defensible than the older "10,000x faster than LangGraph" claim and is the one you should treat as the speed story today. 30+ model providers and 100+ tool integrations sit behind one Python API. The honest catches: the paid Pro tier is the only entry point for a hosted production Control Plane at $150/month with no per-seat or per-connection add-ons published today, and self-hosting the Control Plane is gated to Enterprise. Pick Agno for an Apache 2.0 Python SDK with memory, knowledge, learning, guardrails and a Control Plane you keep inside your own cloud. Look at CrewAI instead if a fully managed cloud runtime matters more than owning the infrastructure yourself.

## How it works

1. Install the SDK (`pip install agno`) and define an Agent, a Team (coordinate, route, broadcast or tasks mode) or a Workflow (Step, Steps, Parallel, Loop, Condition, Router) in Python.
2. Attach the capabilities you need: learning, context providers, memory, knowledge, state, storage, sessions, guardrails, hooks, dependencies, filesystem and human-in-the-loop.
3. Register the same Agents, Teams and Workflows with AgentOS to serve them behind a FastAPI application with 50+ REST endpoints, SSE and websockets.
4. Connect the AgentOS Control Plane (free, local on the developer machine; hosted on Pro; self-hosted on Enterprise) for chat, tracing, sessions, knowledge, memory and evals.

## Who it's for

- Python developers who want a fast, lightweight, model-agnostic framework for building individual agents or collaborating multi-agent teams
- Teams that want built-in memory, knowledge/RAG and native MCP support without assembling those pieces from separate libraries
- Teams that need a production runtime (monitoring, HITL approval, RBAC, audit logs) they operate inside their own cloud account rather than a third-party's

## Strengths and weaknesses

- ✓ Apache 2.0 SDK and runtime, free, with 41,839 GitHub stars and 5,807 forks under agno-agi, an active and growing project.
- ✓ The SDK quotes a 3.2-microsecond agent instantiation and 5.2 KiB of memory per agent, measured on a public harness you can run yourself, a more defensible speed story than the older "10,000x faster than LangGraph" headline figure.
- ✓ The Control Plane connects from the browser directly to a runtime in your own cloud, so sessions, memory, knowledge and traces stay in your database, not a vendor's.
- ✓ Docs ship as an MCP server at docs.agno.com/mcp, so a coding agent (Claude Code, Cursor, Codex) can scaffold an agent platform from documented parts.
- ✗ Self-hosting the Control Plane is gated to Enterprise. Free gets a local-only Control Plane on the developer machine; Pro gets a hosted Control Plane at a real URL but the customer still pays for hosting and operates their own database.
- ✗ The January 2025 rebrand from Phidata and the rename of "Agent Framework" to "Agent SDK" with the 3.0 release mean older tutorials, Stack Overflow answers and blog posts can reference outdated import paths or a pre-3.0 API surface.
- ✗ The Control Plane lives in the same repo as the SDK and AgentOS, so platform upgrades ride along with framework upgrades. A pinned SDK version does not pin the Control Plane.
- ⚠ The 3.2-microsecond / 5.2 KiB figure is measured on Agno's own published harness. Independent, reproducible benchmarks against LangGraph on neutral hardware were not found.
- ⚠ Tutorial content predating the January 2025 Phidata-to-Agno rebrand and predating the SDK 3.0 release may reference a stale import path or API surface.
- ⚠ Agno sends a telemetry event per agent run by default, with prompts, messages and outputs excluded. Disable with AGNO_TELEMETRY=false.

## Key features

- **Apache 2.0 Agent SDK** — Python SDK for building agents, teams and workflows. Three primitives (Agent, Team, Workflow) and attachable capabilities including memory, knowledge, learning, guardrails, hooks and human-in-the-loop.
- **Model-agnostic, 30+ providers** — OpenAI, Anthropic, Google and 27 other providers behind one API. Switch providers without rewriting agent code.
- **100+ tool integrations** — Pre-built toolkits for GitHub, Slack, Postgres, and more. Plus native MCP support, both consuming external MCP servers and exposing agents as MCP servers.
- **AgentOS runtime** — 50+ REST endpoints with SSE and websockets, sessions, memory, knowledge, traces, background execution, evals, scheduler and JWT-based RBAC. Runs in your own cloud.
- **AgentOS Control Plane** — Web UI to manage agents, sessions, knowledge, memory and evals. Connects from your browser directly to your runtime, so data stays in your database. Free locally on the developer machine, hosted on Pro.
- **Docs as an MCP server** — The full documentation is exposed at docs.agno.com/mcp, so a coding agent (Claude Code, Cursor, Codex) can scaffold a platform from documented parts.

## Use cases

- **Building an internal agent platform with coding-agent scaffolding** — A Python team hands the docs MCP URL to Claude Code or Cursor and the coding agent scaffolds an agent platform from documented parts, with memory, knowledge, sessions and a Control Plane the team operates inside their own AWS or GCP account.
- **Multi-agent research or support team** — A developer composes several specialist agents (researcher, writer, reviewer) in an Agno Team that shares context and hands off work, coordinated rather than as a single monolithic agent.
- **Regulated deployment inside a private cloud** — A team whose agents touch sensitive data runs AgentOS in their own Railway, AWS, GCP, Azure or air-gapped Helm environment, so sessions, memory, knowledge and traces stay in infrastructure they control.

## Integrations

30+ LLM providers (OpenAI, Anthropic, Google and others) · 100+ tool integrations including GitHub, Slack, Postgres · Model Context Protocol (MCP), both consuming and exposing · Chat channels: Slack, Telegram, WhatsApp, Discord, AG-UI, A2A · Deploy targets: Docker, Railway, AWS, GCP, Azure, Fly, Render, Modal, Helm

## Sources

- https://www.agno.com
- https://www.agno.com/pricing
- https://www.agno.com/products/sdk
- https://github.com/agno-agi/agno