For the complete documentation index, see llms.txt. This page is also available as Markdown.

Interactive Agent

The Interactive Agent is where agent configurations become running services. Once you have built and saved a configuration this section handles everything needed to bring it to production.

Navigate to Deploy → Interactive Agent to see all agent deployments in your project. The listing shows each agent's name, current status, last update timestamp, and an action button. The top-right corner displays counters for how many agents are deployed and how many are in draft state.

The action button changes depending on the agent's status:

  • Manage appears for agents that are already running. Click it to open the deployment for editing.

  • Deploy appears for agents that have a saved configuration but have not been deployed yet. Click it to open the deployment wizard. You can change the image version, environment variables, secrets, public URL, and schedule.

Deploying an Agent

The deployment wizard opens when you click Save draft and deploy from the Agent Configuration review step, or when you click Deploy on a draft agent in the Interactive Agent listing. The wizard has five steps.

1

Step 1: Identity

Agent configuration shows the build draft this deployment is based on. This field is read-only.

Instance name is the identifier for this deployment. It defaults to the configuration's name and is bumped with a suffix if a deployment with that name already exists. It accepts letters, digits, hyphens, and underscores, and is locked once the agent is live.

Image is the container image that will run the agent. It defaults to interactive-agent, the platform's standard agent image.

Image version is a dropdown listing the available versions of the chosen image, pulled from the operator catalog.

2

Step 2: Endpoint

Public URL is a toggle. When enabled, the agent gets a publicly reachable hostname and a Hostname preview appears showing the full URL where the agent will be accessible. When disabled, the agent stays internal to the platform.

3

Step 3: Runtime

Environment variables are plain, non-secret key-value pairs passed to the agent at runtime. Click + Add variable to define one.

Router API Key and Agent API Key are only shown the first time you deploy an agent.

  • The Router API Key is the outbound key the agent uses to bill AI calls through the platform router. A dropdown lets you pick an existing key from the project or create a new one.

  • The Agent API Key is the inbound key that callers use to authenticate to the deployed agent through HTTP Basic Auth. Click Generate to create one, or enter a value manually. Save it the moment you generate it, because there is no way to reveal it later.

Secrets lists the project secrets available to attach to the deployment. Each selected secret's keys are injected as environment variables at runtime. If no project secrets exist, a message directs you to create one from the Secrets Manager.

4

Step 4: Schedule

Leave the schedule disabled to keep the agent running 24/7. Toggle Enable schedule on to define an uptime window:

  • Uptime accepts a cron-like window expression that defines when the agent should be running. Example: Mon-Fri 09:00-18:00. Outside this window, the agent is scaled to zero to save cost.

  • Timezone accepts an IANA timezone identifier. Example: Europe/Madrid.

5

Step 5: Review

The review screen shows four summary cards (Identity, Endpoint, Runtime, Schedule) alongside a live preview of the Generated manifest in YAML. Each card has an Edit button that jumps back to the corresponding step.

If required fields are missing, a banner at the top reads "Resolve before deploying" and lists the issues with links to the relevant step. Once all requirements are met, click Deploy to publish.

First Deploy

On first deploy with the default image, the platform automatically creates a bundle secret named router-<instance-name>. This secret contains the Router API Key, the Agent API Key, and the platform credentials the agent needs to communicate with the router. The bundle is attached to the deployment automatically, even if you do not select it manually in the Secrets step.

After a successful first deploy, the agent appears in the Interactive Agent listing with a Running status and is reachable at its hostname.

Deployment Statuses

Agents in the listing can show one of three statuses:

  • Running means the agent is live and handling requests at its public endpoint.

  • Draft means the configuration has been saved but not yet deployed. The agent has no running instance.

  • Provisioning means a deployment has been triggered and the platform is bringing the agent up. This status is temporary and transitions to Running once the pod is ready.

Restoring a Previous Deployment

The deployment edit view lists prior revisions. Select one to view it in read-only mode, then use Restore revision to make it live again. No rebuild is needed; the platform redeploys the selected revision as-is.

Last updated

Was this helpful?