> ## Documentation Index
> Fetch the complete documentation index at: https://docs.kugelaudio.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Hosted agents quickstart

> Talk to Juno, KugelAudio's ready-made voice agent, in your browser, then create your own agent, try it on your machine and deploy it. KugelAudio hosts it, with no servers to run.

Put a voice agent on the phone and on your website without running any servers.
KugelAudio hosts the agent, answers its calls and keeps it live. You start by
talking to Juno, a ready-made agent, and then deploy an agent of your own.

<Note>
Hosted agents are switched on per organization. If **Agents** is missing from your
[dashboard](https://kugelaudio.com/dashboard), [ask us](mailto:hello@kugelaudio.com) to
switch it on for yours.
</Note>

<Steps>
  <Step title="Talk to Juno">
    Open **Agents** in the [dashboard](https://kugelaudio.com/dashboard), choose
    **Create agent**, then pick **Juno** under **Add your agent**. Juno is a personal
    assistant that plans your day and remembers notes.

    The dashboard reads Juno's source, builds it and makes it live, then opens the
    **Playground**. Start a call and try "Put buy milk on my list."

    Juno runs on keys KugelAudio provides. Its calls count against your plan's agent
    minutes like any agent's, and the Playground shows how many are left when your plan
    has a limit.
  </Step>
  <Step title="Create your own agent">
    You need Node.js 22.12 or later. Create a project from the agent template. It
    starts as a copy of Juno that you change into your own agent:

    ```bash
    npx --yes --allow-remote=all https://cdn.kugelaudio.com/curb/create-curb-app-latest.tgz my-agent
    ```

    `--allow-remote=all` lets npm 12 run a package from a URL. The project's `AGENTS.md`
    explains how the agent is built: its prompt, skills and tools.
  </Step>
  <Step title="Add your keys">
    The agent needs two keys in the project's `.env` file. The setup asks for them and
    writes `.env`; to add or change one later, edit `.env`.

    - `ANTHROPIC_API_KEY` for the language model, from your Anthropic account.
    - `KUGELAUDIO_API_KEY` for speech, from the [API keys page](/api-reference/authentication).

    The agent hears callers with [Kugel Ear 1](/features/speech-to-text) and answers
with KugelAudio voices, so it needs no other speech key.
  </Step>
  <Step title="Try it on your machine">
    ```bash
    cd my-agent && npm run dev
    ```

    Your browser opens the agent's preview. Type, or press **Start call** and talk.
    The preview shows your agent's context: its instructions, the skills it can open and
    the tools it can call. You do not change the agent in the preview. Ask your coding
    agent to change it, or edit the files yourself, and the preview shows the result.

    Check the project for mistakes, then talk to the agent in the terminal, the way a
    coding agent does:

    ```bash
    npx curb check
    npx curb say "Hello"
    npx curb test
    ```

    `curb say` runs one caller turn and the next one continues the conversation.
    `curb test` replays the conversations kept in `src/**/*.scenario.ts`, with scripted
    caller lines or a simulated caller, and fails when one goes wrong.
  </Step>
  <Step title="Deploy">
    ```bash
    npx curb login
    npx curb secrets push
    npx curb deploy
    ```

    `curb login` opens the KugelAudio sign-in page in your browser; confirm, and the
    terminal is signed in. `curb secrets push` sends the keys in `.env` to the agent
    and prints their names, never the values. It skips `KUGELAUDIO_API_KEY`, which
    KugelAudio provides to every hosted agent. `curb deploy` uploads the project,
    builds it and makes it the agent's live version. It ends with a link to the agent's
    Playground, where you can call it.
  </Step>
</Steps>

Calls are billed per minute; see [pricing](https://kugelaudio.com/pricing).

## Let your coding agent do it

**Build your own agent** on Juno's Playground has a prompt for a coding agent such as
Claude Code, Cursor or Codex. Start the coding agent in an empty folder on your machine and
paste the prompt: it asks what your agent should do, then runs the steps above for you. It
asks you for your keys and never prints them.
[Connect it to KugelAudio](/hosted-agents/coding-agent) first, and it also deploys and reads
build and call errors through the MCP server.

## Next steps

<CardGroup cols={2}>
  <Card title="Connect your coding agent" icon="code" href="/hosted-agents/coding-agent">
    Let Claude Code build, deploy and debug the agent with you
  </Card>
  <Card title="Recipes" icon="book-open" href="/hosted-agents/recipes">
    Call links, recording notices, feedback, your own APIs, test copies
  </Card>
  <Card title="Deploy from your terminal" icon="terminal" href="/hosted-agents/deploy">
    Every option of `curb login`, `curb secrets push` and `curb deploy`
  </Card>
  <Card title="Deploy on git push" icon="github" href="/hosted-agents/github">
    Connect a GitHub repository and ship every push
  </Card>
  <Card title="Web widget" icon="globe" href="/hosted-agents/web-widget">
    Put a talk button on your website
  </Card>
  <Card title="Phone numbers" icon="phone" href="/hosted-agents/phone-numbers">
    Buy a number, or answer the numbers you already have
  </Card>
</CardGroup>