Deploy an AI agent

This tutorial takes you from a signed-in account to a running AI agent with a chat UI, a spending limit, and a clean shutdown. It takes about ten minutes. You will use the Alternate Agent template, which bills its model usage to your Alternate Clouds credits, so no third-party AI keys are needed.

The agent is a composite template (the agent plus its own PostgreSQL database), and composite templates deploy from the web app only; the CLI does not support them yet. Everything after the deploy works from the CLI.

Before you start

• You have finished the Quick start: acc is installed, you are logged in, and a project is active (acc whoami shows it).
• You are signed in to the web app at https://clouds.alternatefutures.ai with the same account.
• Your organization has credits or an active trial.

1. Look at the template

acc templates info alternate-agent

The output lists the resources (1 vCPU, 2 GiB memory, 2 GiB storage), the port, and the environment variables. Two of them are filled in by the platform when you deploy:

Everything else has a default. You will set AGENT_NAME and AGENT_INSTRUCTIONS to make it yours.

2. Deploy from the web app with a budget cap

1. Open Deploy a template at https://clouds.alternatefutures.ai/deploy and choose Alternate Agent.
2. Set AGENT_NAME (for example Atlas) and AGENT_INSTRUCTIONS (for example You are a concise assistant for the Atlas project.). Leave AF_API_KEY and AF_ORG_ID as the platform pre-fills them.
3. Under Budget Cap, set a Monthly limit of 20. The platform stops the service once it has spent 20 dollars in a month; your machine is not involved.
4. Deploy.

The template creates two services in your project, the agent and its database. The deployment now goes through placement and start; How a deployment works explains each stage.

3. Watch it come up

acc services list

Run it again every twenty seconds or so until the agent service shows as active with a URL of the form https://my-agent-app.alternatefutures.ai. If it is still waiting after a few minutes, see The deployment stays in a waiting or creating state.

To see the agent starting, use the id acc services list printed:

acc services logs <serviceId> --tail 100

4. Talk to it

Open the URL from step 3 in a browser. The chat UI loads with the name you set. Ask it something; the reply is billed to your credits through the token the platform created for it, and you can see the charge later under Billing, Usage in the web app.

5. Change it without redeploying by hand

Environment variables live on the service. To give the agent a different personality:

acc services env set <serviceId> AGENT_INSTRUCTIONS "You answer only with haiku."
acc services deploy <serviceId>

The redeploy creates a new deployment; the URL does not change.

6. Shut it down

Close both services the template created, so nothing keeps billing:

acc services list
acc services close <serviceId>   # once for the agent, once for its database

acc services close keeps the service definition so you can deploy it again later. acc services delete <serviceId> removes it entirely.

What you learned

• Templates carry their own resources and declare the variables they need; some variables the platform fills in for you.
• Composite templates deploy from the web app; from then on the CLI manages their services like any other.
• A per-service budget cap is set at deploy time and enforced server-side.
• Configuration changes are environment variables plus a redeploy.
• Closing a service is how you stop paying for it.

Next steps

• How a deployment works
• How billing works
• Template catalog, every template with its variables
• Docs for AI agents, for running the CLI steps from CI or another agent