> ## Documentation Index
> Fetch the complete documentation index at: https://agno-v2-docs-align-with-readme.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Deploy

> Ship AgentOS as a normal Python service. Container, Postgres, public hostname.

AgentOS is a FastAPI app. Deploy it like any normal Python service: a container, a Postgres database, a public hostname, and a few env vars.

```python theme={null}
from agno.os import AgentOS

agent_os = AgentOS(agents=[agent], db=db, tracing=True)
app = agent_os.get_app()

if __name__ == "__main__":
    agent_os.serve(app="my_app:app", reload=False)
```

## What you need to ship

| Resource            | Why                                                                                    |
| ------------------- | -------------------------------------------------------------------------------------- |
| **Container host**  | Runs the FastAPI process                                                               |
| **PostgreSQL**      | Sessions, memory, knowledge, traces, schedules                                         |
| **Public hostname** | Required for Slack, Telegram, WhatsApp interfaces                                      |
| **HTTPS**           | Required for every webhook interface; terminate at your load balancer or reverse proxy |
| **Env vars**        | At minimum `OPENAI_API_KEY` and `JWT_VERIFICATION_KEY` (in prod)                       |

AgentOS handles queues, worker pools, the scheduler, and JWT auth in-process. No separate worker fleet, no separate auth server, no separate cron container.

## Local with Docker Compose

```yaml theme={null}
services:
  agentos:
    build: .
    ports:
      - "8000:8000"
    env_file: .env
    depends_on:
      - db

  db:
    image: agnohq/pgvector:18
    environment:
      POSTGRES_USER: ai
      POSTGRES_PASSWORD: ai
      POSTGRES_DB: ai
    volumes:
      - pgdata:/var/lib/postgresql/data

volumes:
  pgdata:
```

`agnohq/pgvector` is Postgres 18 with the pgvector extension preinstalled — needed for knowledge embeddings.

```bash theme={null}
docker compose up -d --build
curl http://localhost:8000/health
```

The [Scout](/tutorials/scout/setup), [Dash](/tutorials/dash/setup), and [Coda](/tutorials/coda/setup) tutorials all start here.

## Railway

Each tutorial template ships with Railway scripts for one-command deploys:

```bash theme={null}
railway login
./scripts/railway_up.sh        # provision Postgres + app, get a domain
./scripts/railway_env.sh       # sync .env.production to the service
./scripts/railway_redeploy.sh  # push code updates
```

`up.sh` provisions the project, adds `pgvector` with a persistent volume, creates the app service with env vars, and assigns a public domain.

Walkthroughs: [Scout deploy](/tutorials/scout/deploy-to-railway), [Dash deploy](/tutorials/dash/deploy-to-railway), [Coda deploy](/tutorials/coda/deploy-to-railway).

## AWS, GCP, and Azure

Any container platform works. The shape:

| Component                   | Service options                                | What runs in it                          |
| --------------------------- | ---------------------------------------------- | ---------------------------------------- |
| **App service**             | ECS Fargate, Cloud Run, App Service            | The AgentOS container, port 8000         |
| **Postgres**                | RDS, Cloud SQL, Postgres Flexible Server       | Sessions, memory, knowledge, traces      |
| **Load balancer / ingress** | ALB, Cloud Load Balancing, Application Gateway | Public HTTPS termination                 |
| **Secret manager**          | Secrets Manager, Secret Manager, Key Vault     | `OPENAI_API_KEY`, `JWT_VERIFICATION_KEY` |

Health check the `/health` endpoint. AgentOS responds `{"status":"ok"}` when the app is ready.

## Scaling

AgentOS is stateless. State lives in `db`. Scale horizontally:

| Concern                                    | Solution                                                                                                           |
| ------------------------------------------ | ------------------------------------------------------------------------------------------------------------------ |
| Throughput                                 | Add app replicas behind a load balancer                                                                            |
| LLM rate limits                            | Use a queue or rate limiter in front of the model client                                                           |
| Long-running runs                          | Use `background=true` on the run endpoint, then poll for completion (see [Serve as an API](/runtime/serve-as-api)) |
| Side effects without blocking the response | Background hooks with `run_in_background=True`                                                                     |
| Schedule fan-out                           | The scheduler runs on a single replica's lifespan; for HA, use leader election or pin scheduling to one replica    |
| Trace volume                               | Use a separate `trace_db` to keep the primary lean (see [Observability](/runtime/observability))                   |

For the leader-election pattern with multiple replicas, see [Scheduler HA](/agent-os/scheduler/overview).

## Production checklist

**Auth and secrets**

* `RUNTIME_ENV=prd` enables JWT auth
* `JWT_VERIFICATION_KEY` set (see [Security & Auth](/runtime/security-and-auth))
* `OPENAI_API_KEY` and other model keys in a secret manager, not in source

**Infrastructure**

* Postgres has a persistent volume or managed backup
* HTTPS terminating at your load balancer or reverse proxy
* Health check pointed at `/health`

**Operational**

* Tracing on (`tracing=True`) so you can debug bad runs
* At least one [interface](/runtime/interfaces) wired up
* [Pre-hooks](/runtime/human-approval) for PII or injection guarding if you handle untrusted input
* `requires_confirmation=True` on irreversible tools

## Updating your deployment

Code changes: `git push` if CI auto-deploys, or `./scripts/railway_redeploy.sh`. Env changes: `./scripts/railway_env.sh` (Railway auto-redeploys when env values change). Database changes: AgentOS handles its own tables — schema changes are additive and forward-compatible, so no migration tool is required for stock AgentOS tables. Application tables you migrate however you like (Alembic, raw SQL, dbt, your call).

## Next

[Build a Product →](/runtime/build-a-product)