You will take the my-investigator project from the quickstart and run it on a managed environment. Same code, same workflow — only the worker host changes from your laptop to AGNT5’s managed runtime.

Time: about 5 minutes.

You’ll learn:

• Push your model provider key as a project-scoped secret
• Build and ship the project to a managed environment
• Trigger the deployed workflow and inspect its trace

Prerequisites:

• The my-investigator project from the quickstart, authenticated with agnt5 auth login.

Step 1: Push the OpenAI key as a secret

Local .env files do not travel to the cloud. From inside the project directory:

agnt5 secrets set --name OPENAI_API_KEY --type api_key

The CLI prompts for the value and stores it scoped to this project. Confirm:

agnt5 secrets list

If you also configured an alternative search provider (AGNT5_BRAVE_SEARCH_API_KEY, AGNT5_TAVILY_API_KEY, or AGNT5_SEARXNG_URL), push that secret the same way. The provider-hosted built-in path needs only OPENAI_API_KEY.

Step 2: Deploy

agnt5 deploy

agnt5 deploy defaults to the preview environment. The command:

1. Builds your project image with a managed BuildKit instance.
2. Pushes the image to the registry.
3. Creates a deployment record.
4. Provisions workers and waits until they report ready.

Output streams progress in real time. When it finishes you get a deployment ID:

✓ Build complete
✓ Image pushed
✓ Deployment created: d8f3a2b1-1234-5678-9abc-def012345678
✓ Workers ready (1/1)

Step 3: Verify the deployment

agnt5 deploy status <deployment-id>

The output names the environment, the replica count, and the worker health. Pass --watch to refresh every two seconds.

To list every deployment in the project:

agnt5 deployments

Step 4: Trigger the workflow remotely

agnt5 run investigate_with_review --env preview --input '{
  "question": "Should we migrate from Redis to Valkey?"
}'

Without --env preview the CLI routes to your active dev session. With it, the request hits the deployed environment.

The workflow runs the same path as it did locally:

1. Connects to DeepWiki over Streamable HTTP.
2. Drafts a brief.
3. Pauses for human review.

Step 5: Approve in Studio

Open Studio (default app.agnt5.com; agnt5 context show for custom contexts). Switch the project’s environment selector from dev to preview. The deployed run shows up at the top of the runs list, paused at the human review step. Approve, edit, or reject as you would in dev.

After approval, the workflow saves the report on the managed worker. The file lands inside the worker container — see the deployment’s storage configuration for how to retrieve it.

Step 6: Inspect the deployed run from the CLI

agnt5 inspect runs --env preview
agnt5 inspect trace -r <run-id>

The trace shape matches your dev run: the same MCP step, the same agent loop, the same wait_for_user, the same save_report. Workers changed; the workflow did not.

What you built

The same workflow you ran in your dev session is now running under managed workers. Updates flow with another agnt5 deploy. Promote preview to staging or production by passing --env.

What you did not write or configure:

• A container build pipeline
• Image registry credentials
• Worker process supervision and health checks
• Secret distribution to workers

Next steps

• Improve with evals — capture deployed runs as eval data and measure changes before you redeploy.
• Workflows — the durable-execution model that lets the same code run unchanged in dev and cloud.