agno-agi/agno

A Python framework and runtime for building, running, and managing agent platforms you host yourself.

What it is

Agno is a framework and runtime for agent platforms. You build agents with the Agno SDK, run them as a service through the AgentOS runtime, and manage the platform through the AgentOS web UI. It is aimed at developers who want to own their agent stack — keeping data, memory, and security posture under their own control rather than delegating them to a hosted product.

Key features

Production API with 50+ endpoints, plus SSE and websocket support, for building a product on top of your agents.
Storage for sessions, memory, knowledge, and traces in your own database.
100+ prebuilt toolkit integrations (GitHub, Slack, Postgres, and others) and context providers that pull live data from Slack, Drive, wikis, MCP, and custom sources.
Human-in-the-loop approval that can pause runs for confirmation and block tools requiring admin sign-off.
Observability through OpenTelemetry tracing, run history, and audit logs.
JWT-based RBAC with multi-user, multi-tenant isolation, plus interfaces for Slack, Telegram, WhatsApp, Discord, AG-UI, and A2A.

Tech stack

Primary language: Python.
AgentOS runtime typically deployed with Docker, backed by a Postgres database for data and traces, and packaged with an MCP server and a control plane.
Observability via OpenTelemetry.
Starter templates for Railway, Docker, AWS, GCP, Azure, Fly, Render, Modal, and Helm.
License: Apache-2.0. No root dependency manifest was included in the inputs, so specific version constraints are not available here.

When to reach for it

Standing up a self-hosted agent platform where you need a REST API, storage, and a management UI rather than a single script.
Serving multiple users or tenants that require RBAC and isolation out of the box.
Building products that need approval gates, scheduling, and audit logging around agent runs.
Deploying the same agent stack across different cloud providers using the provided starter templates.

When not to reach for it

A quick one-off script or notebook experiment, where a full runtime and control plane are more infrastructure than the task warrants.
Teams outside the Python ecosystem, since the SDK is Python-first.
Cases where you specifically want a fully managed, hosted service and do not want to run and maintain the runtime and database yourself.

Maturity signal

The project has been under development since 2022 and shows a very recent push date, which points to active, ongoing maintenance rather than a dormant or abandoned codebase. A large star count and a substantial open-issue backlog are consistent with a widely used project that is still moving quickly and fielding real-world usage. The Apache-2.0 license is permissive and business-friendly, and the repository is not archived.

Alternatives

LangChain / LangGraph — reach for it when you want a broad, established library ecosystem and graph-based orchestration rather than an opinionated hosted runtime.
CrewAI — reach for it when your focus is multi-agent role/crew coordination and you do not need a full platform layer.
LlamaIndex — reach for it when the core problem is data ingestion and retrieval over your documents rather than serving and managing an agent platform.

Notes

The recommended "get started" path hands a prompt to a coding agent (Claude Code, Cursor, Codex) that clones a starter repo and follows its guide, rather than a manual setup.
Agno sends a telemetry event per agent run to prioritize model providers; it states prompts, messages, and outputs are never sent, and telemetry can be disabled with AGNO_TELEMETRY=false.

Wiki: agno-agi/agno