Recipes for running Agentgram outside your laptop. Pick the one that matches where you want to host it.
| Recipe | Best for | What you get |
|---|---|---|
docker-compose/ |
A single VM or server | API + web + Redis + PostgreSQL, using the published images |
Root docker-compose.yaml |
Trying it in two minutes | Same stack; baked-in demo agent |
all-in-one/ |
Fewer app containers | One agentgram image (API+UI) + Redis + PostgreSQL |
kubernetes/ |
A cluster | Helm values for the bjw-s app-template chart + an Ingress |
Both pull the official images from GitHub Container Registry:
ghcr.io/dfradehubs/agentgram:<version>
ghcr.io/dfradehubs/agentgram-api:<version>
ghcr.io/dfradehubs/agentgram-web:<version>
Use a pinned version tag (e.g. v0.1.0) in production instead of latest.
Regardless of where you run it, the API depends on:
- PostgreSQL — agents, permissions, metrics and persistent data.
- Redis — sessions and pub/sub for live streaming.
- A config file (
CONFIG_PATH) — every secret is read from the environment via${ENV:VAR}.
Database migrations run automatically on API startup, so there is no separate migration step.
The examples ship with auth.enabled: false so you can try Agentgram immediately. Do not expose an
unauthenticated instance to the public internet. Before going live, enable basic auth or generic
OIDC in the config. See the root README and api/docs/SECURITY.md.