Your agent in 5 minutes

You have an agent that already works on your laptop and you want it running in your AWS account, isolated in a MicroVM and with its model credential out of its reach. This page does it with one command and no changes to its code. The details of each step are in Deploy your agent.

What you need

• Rayito set up in your account, with the rayito-base-caps image published (rayito setup --caps-image).
• A Bedrock API key in the AWS_BEARER_TOKEN_BEDROCK variable (or your provider's: ANTHROPIC_API_KEY, OPENAI_API_KEY…).
• An agent project with its lockfile (uv.lock, requirements.txt or package-lock.json) and the model id in the code. A LangGraph graph, a Strands agent, a deepagents agent, a Vercel AI SDK agent, a Hermes agent or an HTTP app will do.

1. See what it detects

rayito agent detect ./mi-agente

It reads the project files; it does not run your code or call AWS. If something is not what you expected, change it with --framework, --entry or by writing the key in a rayito-agent.toml (--write saves it for you to edit).

2. See the plan and the cost

rayito agent deploy ./mi-agente --dry-run --max-sessions 2 --min-vms 1 --max-vms 2 --warm 1 --memory 4096

It creates nothing. It tells you which bucket, secret and base image it would use, how many VMs you would have and how much each costs per hour.

3. Deploy it

export AWS_BEARER_TOKEN_BEDROCK=...        # your key; it is stored in Secrets Manager and not written to disk
rayito agent deploy ./mi-agente --create-secret

The first time it builds the agent image (a few minutes: it installs the runtime and your dependencies with their pinned versions); later times reuse it if nothing changed. When done it prints the fleet name and leaves the access token in .rayito/agent-token (mode 0600; do not commit it).

4. Call it

CLIPythonTypeScript

rayito agent invoke mi-agente "What is 17 plus 25?"

from pathlib import Path

from rayito import AgentFleet

fleet = AgentFleet.connect(
    "mi-agente", access_token=Path(".rayito/agent-token").read_text().strip()
)
reply = fleet.invoke("What is 17 plus 25?")
print(reply.text)

import { readFileSync } from "node:fs";
import { AgentFleet } from "rayito";

const fleet = await AgentFleet.connect("mi-agente", {
  accessToken: readFileSync(".rayito/agent-token", "utf8").trim(),
});
const reply = await fleet.invoke("What is 17 plus 25?");
console.log(reply.text);

5. Stop it

rayito agent stop mi-agente

It terminates every VM of the fleet. If you do not stop it, it lives at most 8 hours and bills while it runs; with --idle 300 it suspends itself when nobody uses it and bills no compute.

Next

• Several sessions and several VMs: more than one conversation at a time, queue, warm slots and what happens if the app dies.
• Deploy your agent: the project file, the frameworks and how the credential reaches the model.
• Deploy to AgentCore: the same project on Amazon Bedrock AgentCore Runtime with --target agentcore, without local Docker (the image is built in CodeBuild).
• Deploy with CDK: the agent as a RayitoAgent construct in your CDK app, deployed with cdk deploy.