AgentOS is an open-source agent platform that runs in your cloud. It comes with a Super Agent that builds agents, coordinates their work, and helps you manage the platform. Ask it to run an agent, create a workflow, or break a larger goal into Projects and Tasks.
Built with Agno, AgentOS brings together:
- Agents, teams, and workflows you build in Python, Studio, or through Super Agent.
- Memory, knowledge, and storage backed by your own PostgreSQL database.
- Permissions, human approvals, and scheduling for managing work.
- An API and MCP server to use your agents in products, Claude, and ChatGPT.
Use the AgentOS UI to chat with agents, inspect runs and traces, and track their work. This template runs locally with Docker and deploys to Azure Container Apps.
Give this prompt to your coding agent:
Help me set up my agent platform and build my first agent.
Clone https://github.com/agno-agi/agentos-azure into a folder called agent-platform,
cd into it, and follow .agents/skills/setup-platform/SKILL.md.
It checks Docker, helps you configure your API key, starts the platform, and builds
an agent with you. If you've already cloned the repo, ask it to run /setup-platform.
You'll need Git, Docker, and an OpenAI API key.
git clone https://github.com/agno-agi/agentos-azure agent-platform
cd agent-platform
cp example.env .envSet OPENAI_API_KEY in .env, then start the platform:
docker compose up -d --buildOnce localhost:8000/docs loads:
- Sign in at os.agno.com.
- Choose Connect OS, select a local connection, and enter
http://localhost:8000. - Open Super Agent and tell it what you're working on.
Start with a concrete request:
Build an agent that turns meeting notes into decisions and action items, then run it on these notes: …
Super Agent can use your existing agents or build one for the job. Agents it builds appear in the Agents list and are available through the API and MCP.
For ongoing work, ask it to create a Project:
Propose a Project to build a customer support agent. Break it into Tasks for my review.
The Company page keeps track of Products, Projects, Tasks, and scheduled Routines. You can approve work, assign it, review results, and request changes there or through Super Agent. Products describe ongoing responsibilities; Projects and Tasks organize specific work; Routines run on a schedule.
This template includes a weekly maintenance Routine that reviews the platform for relevant Agno updates, security issues, and quality problems. It is disabled by default. Enable it from the Company page to receive proposals for review.
This repo includes skills for building and managing your agent platform. Ask your coding agent to follow the relevant skill:
| Task | Skill |
|---|---|
| Build an agent | /create-agent |
| Add tools, change behavior, or fix a bug | /extend-agent |
| Write evals | /create-evals |
| Find weaknesses and improve reliability | /improve |
| Review the repository before release | /review-and-improve |
This repo comes with Claude Code and Codex agents. Work with them directly in AgentOS or ask Super Agent to delegate work to them.
They're off by default. Follow the harness setup guide to enable them and configure credentials and sandboxes.
Once enabled, ask Super Agent:
Ask Codex Engineer to review this project and suggest improvements. Don't change any files yet.
Without a sandbox, engineers run on the API host with its file and credential access. Sandboxes provide isolated execution and checkpoint recovery.
Create an empty private GitHub repository, then point this checkout at it:
git remote rename origin upstream
git remote add origin https://github.com/YOUR-ORG/YOUR-REPO.git
git push -u origin HEADYour work now goes to your repository; upstream keeps the template available for
updates. If you use sandboxes, also set your repository URL and branch in the
harness configuration.
Ask your coding agent to follow
/deploy-platform. This template targets Azure Container Apps.
You’ll need Azure CLI (az login), Docker and OpenSSL.
Set AZURE_RESOURCE_GROUP and AZURE_LOCATION; defaults are agentos and eastus. Export a custom resource group for later sync/redeploy/teardown commands. Setup registers providers, checks PostgreSQL region restrictions and reuses existing resources. Environment sync creates a fresh revision so changed secrets take effect. See the Azure troubleshooting guide.
Create .env.production from example.env if absent. Set OPENAI_API_KEY
and any other model credentials you need. Keep RUNTIME_ENV=prd or leave it unset.
./scripts/azure/up.shWhen setup provides the HTTPS URL, connect it as a Live connection at
os.agno.com. Enable JWT authorization in Settings → OS Security
and supply the public verification key. If the prompt was skipped, save the full quoted
multiline JWT_VERIFICATION_KEY in .env.production, then run
./scripts/azure/env-sync.sh. The API requires the key to start in production.
Use ./scripts/azure/redeploy.sh for code and ./scripts/azure/env-sync.sh for settings.
Keep AGENTOS_URL set to the public HTTPS URL for scheduling and MCP OAuth. Verify
/health, an authenticated UI connection, and rejection of anonymous /agents requests.
Run one API instance and read scaling before adding replicas.
For optional hosted sandboxes, follow harness setup and preserve
SANDBOX_PROVIDER=agentcore plus the required AWS credentials, runtime and S3 settings.
For MCP clients like Claude and ChatGPT, set MCP_CONNECT_SECRET to a random secret in your production environment and apply it with the settings update command above. Connect the client to https://YOUR-DOMAIN/mcp and enter the secret on
the consent page.
For supported coding clients:
uvx agno connect --url https://YOUR-DOMAINInstall uv, then create and activate a virtual environment
from the repository root. If another environment is active, run deactivate first.
./scripts/venv_setup.sh
source .venv/bin/activateThe setup script installs the application and development dependencies into .venv.
After making changes, format your code and check formatting, lint, and types:
./scripts/format.sh
./scripts/validate.shWith PostgreSQL running, test Super Agent with the smoke evals. These spend model calls and save results in the database:
python -m evals --tag smokeAdd cases in evals/cases.py. Use ./scripts/mcp_check.sh to check
the running MCP server without a model call.
See AGENTS.md for development conventions and Agno documentation for framework guides.