Are you an LLM? Read llms.txt for a summary of the docs, or llms-full.txt for the full context.
Skip to content

Source & Deploy In Amodal

Use Amodal to create an agent, deploy its source, and inspect the result. For example source to deploy, follow the support-agent tutorial.

1. Choose The Source

Select your organization and create an agent. Choose one of these paths:

  • Template: select a template and name the agent. Amodal stores its source and opens Deploys.
  • Blank agent: select Blank agent and name it. Its source is an amodal.json with only a name and a version. Amodal stores it and opens Deploys. Tick Let the agent propose changes to itself to also add agents/main/AGENT.md and an agents/main/agent.json that declares list_repo_files, read_repo_file, and propose_source_change. The agent can then propose changes to its own prompt, skills, knowledge, and tool code when asked in chat. Each change waits under Proposals until someone approves it. See Proposals.
  • Git repository: choose Connect a git repository instead, select the repository, and name the agent. GitHub uses the Amodal GitHub App. GitLab is available when enabled for the platform.

For GitHub, the App installation determines which repositories appear. If yours is missing, use Configure GitHub App permissions to grant access.

Template and blank agent source can be exported to an empty GitHub repository from Settings > Source. You can also keep editing it in Amodal.

2. Configure The Agent

Before deploying, check:

  • Models: choose the model. The hosted default is Gemini 3.5 Flash.
  • Organization settings: configure provider credentials and allowed models when needed.
  • Secrets: set connection credentials and other values referenced with env:NAME.
  • Auth, embed, and domains: configure app access and embedding.
  • Runtime and deploy settings: review runtime resources, source settings, and automatic deployment behavior.

The model choice and secret values are platform settings. Keep secret references in source; keep the values in Amodal.

3. Deploy And Select Production

Open Deploys, select the source version, and deploy it. Wait for the build to finish, then check the deploy's runtime status and URL.

Source versions and deployments have separate histories. One source version can have several deployments. With manual production promotion, the default setting, a completed production build stays inactive. Automatic deployment can start a build without making it production.

Open the deploy URL to test that version, then use Make production and check the production indicator. The persistent agent URL serves the active production deployment. See Deployment API for API activation, promotion, and rollback.

4. Chat And Inspect

Ask a question that exercises one of the agent's connections or tools. In the support tutorial, use:

Triage ticket 1. What should the support operator do next?

Inspect the messages and tool calls in Sessions, and the model, token usage, and timing in Traces. Check Logs for runtime errors. Before the first deployment, runtime pages may have no data.

5. Edit, Test, And Redeploy

Edit the source files and review the diff before committing. For a connected Git repository, publish the changes to the branch you intend to deploy. For Amodal-hosted source, save a source version in Amodal.

Deploy the changed source, run the agent's evals, and test the deploy URL. Select production after checking the result.