Skip to content
← Tools / Skills
ToolAgno (formerly Phidata)/10 trustVerified
Agno logo

Agno

Build, run, and manage agent platforms

Visit official site →

About

Agno (formerly Phidata) is an open-source, model-agnostic framework for building multi-agent systems with memory, knowledge, and reasoning. It provides an SDK for agents, multi-agent teams, and stateful workflows, plus the AgentOS runtime — a FastAPI backend that serves agents in production with sessions, tracing, scheduling, and JWT-based RBAC. The AgentOS UI control plane (os.agno.com) connects directly from your browser to your self-hosted runtime, so data stays in your own infrastructure. The project ships 100+ pre-built toolkits/integrations and starter templates for deploying on Railway, Docker, AWS, GCP, and others.

What it can do

  • • Agents, multi-agent Teams (route/broadcast/tasks coordination), and stateful Workflows (Step, Parallel, Condition, Loop, Router)
  • • Memory: user profiles, entity memory, session context; knowledge for RAG over documents
  • • Reasoning modes and guardrails; human-in-the-loop approvals and tool gating
  • • 100+ pre-built toolkits/integrations (GitHub, Slack, Postgres, etc.) and context providers (MCP, wikis, Drive)
  • • AgentOS runtime: FastAPI backend with SSE/websocket endpoints, tracing, cron scheduling, JWT-based RBAC
  • • Interfaces: Slack, Telegram, WhatsApp, Discord, AG-UI, A2A protocol
  • • AgentOS UI control plane that connects browser-directly to your self-hosted runtime
  • • Deploy-anywhere starter templates: Railway, Docker, AWS, GCP, Azure, Fly, Render, Modal, Helm

Setup

  1. Create a project and install (per official quickstart): mkdir personal_agent && cd personal_agent, uv init --bare --python 3.14.
  2. Add dependencies: uv add "agno[os,sqlite]" openai.
  3. Export your model key: export OPENAI_API_KEY=<key> (Agno is model-agnostic; other providers supported).
  4. Write your agent: Agent(name=..., model="openai:gpt-5.6", db=SqliteDb(...), tools=[...]), wrap in AgentOS(agents=[agent], db=db, tracing=True) and serve with agent_os.serve(app="personal_agent:app").
  5. Run: uv run personal_agent.py — the API is live at http://localhost:7777.
  6. (Optional) Connect the AgentOS UI: open os.agno.com → Connect Existing OS → Local → http://localhost:7777.
  7. (Privacy) Disable telemetry: export AGNO_TELEMETRY=false — the README discloses a per-run telemetry event (no prompts/outputs sent).
  8. For production: deploy an AgentOS starter template (e.g. agentos-railway, agentos-docker, agentos-aws) instead of local SQLite.

Prerequisites

Python 3.10+ (uv recommended); an LLM API key (OpenAI or any supported provider); Docker for the AgentOS starter templates; Postgres for production deployments.

Use cases

chatautomationreasoningdata

Works with

Claude CodeCursorCodexWindsurfGeneric APIAny agent (.md)

Trust score — 9/10

/10 trustVerified

Scored 9/10 (verified tier, metrics-v1). Evidence: 42,686 GitHub stars (log-scaled vs catalog peers); blended with ~1,779,568 monthly package downloads; last push 0d ago; 100 releases/12mo; 1892 open issues. Manual review: default-on telemetry (opt-out disclosed); 1,892 open issues; post-rebrand API churn risk. License: Apache-2.0.

★ 42,686 GitHub stars

How the trust score is built

Full methodology →
Adoption · stars/downloads, log-scaled vs peers
8.7/10
Maintenance · recency, releases, issue load
8.2/10
Openness · license + docs
10.0/10
Manual review · sanity read + red-flag check
8.0/10
  • · 42,686 GitHub stars (log-scaled vs catalog peers)
  • · blended with ~1,779,568 monthly package downloads
  • · last push 0d ago; 100 releases/12mo; 1892 open issues
  • · manual review: default-on telemetry (opt-out disclosed); 1,892 open issues; post-rebrand API churn risk

Sources

Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.