Integrations overview
Use an integration to choose where Agents run, issue ChatGPT service-account credentials, give an Agent repository access, or connect it to Slack. Installation operators select Drivers and experimental Backends; people configuring an Agent select its channel and model authentication.
| If you want to… | Start here |
|---|---|
| Run Agents on Kubernetes or an existing Linux host | Drivers overview |
| Choose which Drivers an Installation loads | Drivers quickstart |
| Issue managed ChatGPT credentials for dedicated Codex Agents | ChatGPT Backend (experimental) |
| Let people send messages to an Agent in Slack | Slack |
| Give an Agent access to GitHub repositories | Repository access |
Backends (experimental) lists what an Installation can configure. A ChatGPT Backend manages service accounts; it does not route inference. To give an Agent an existing OpenAI API key, use the Agent's model authentication instead.
Kubernetes Compute is the recommended starting point for new Agents. The Docker Compute Driver used by Docker/Podman Compose is limited to control-plane development; it cannot deploy new Agents under the current authentication contract. This limit does not apply to Kubernetes running in local Docker through Local Setup. OpenShell is documented under Drivers with its upstream blockers; it is not supported for production.