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

# Agent skill

> Set up and use Tazo with Codex, Claude Code, Cursor, or another coding agent.

The Tazo agent skill gives your coding agent instructions, SDK examples, and
verification helpers for making decisions from audio and text prompts. Use it
to create an integration, add streaming, or check that your setup works.

The skill is available in the public
[tazolabs/agent-skills repository](https://github.com/tazolabs/agent-skills).

## Prerequisites

* **Node.js 22.20+** to run the skill installer.
* **Python 3.11+** to use the Tazo Python SDK and bundled helpers.
* A Tazo API key from the [Tazo console](https://app.tazolabs.com) for live
  decisions and model discovery. See [Authentication](https://docs.tazolabs.com/authentication.md).

The agent can install `tazo-sdk` in your project's Python environment. Installing
the skill itself does not install the Python package or create an API key.

## Install the skill

Run the installer from the project where you want to use Tazo. Select your
coding agent when prompted, or choose it explicitly:

```bash Choose an agent theme={null}
npx skills@latest add tazolabs/agent-skills --skill tazo
```

```bash Codex theme={null}
npx skills@latest add tazolabs/agent-skills --skill tazo --agent codex
```

```bash Claude Code theme={null}
npx skills@latest add tazolabs/agent-skills --skill tazo --agent claude-code
```

```bash Cursor theme={null}
npx skills@latest add tazolabs/agent-skills --skill tazo --agent cursor
```

Installation applies to the current project by default. Add `--global` to make
the skill available across your projects. Reload your agent's session if it
does not discover the new skill.

To share the skill manually, share the entire
[`skills/tazo` folder](https://github.com/tazolabs/agent-skills/tree/main/skills/tazo),
including its references, scripts, and sample audio.

## Configure your key

Set the complete key, including its `tazo_sk_` prefix, as `TAZO_API_KEY` in your
terminal or secret store. Use the raw key without adding `Bearer `. The SDK
adds the authorization prefix for you.

Keep the key out of source files and agent chat. Start your agent in an
environment where it can access the variable, or configure it through your
agent's supported secret mechanism.

The default API origin is `https://api.tazolabs.com`. If you use another
deployment, tell your agent its API origin and configure a key valid there.

## Ask your agent to use Tazo

Mention the Tazo skill and describe the recording, decision, and desired mode.
In Codex, you can explicitly invoke the skill with `$tazo`.

### Make a decision from a recording

> Use the Tazo skill to set up this Python project and make a decision from
> `request.wav`. Route weather requests to `get_weather`, otherwise `NO_TOOL`.
> Use a synchronous upload and print the selected choice and probabilities.

The agent can discover ready models, check their capabilities, and create a
script using the [Python SDK](https://docs.tazolabs.com/sdk/python/quickstart.md). Tazo returns the selected
decision; your application decides whether to execute a tool named in a choice.

### Add streaming decisions

> Use the Tazo skill to add async streaming decisions to this application.
> Discover the available models, use one choice question, and send audio while
> receiving decision events. Handle the final result and close the session.

The skill covers PCM audio requirements, concurrent sending and receiving,
negotiated limits, and session cleanup. See [SDK streaming](https://docs.tazolabs.com/sdk/python/streaming.md)
for the underlying client interface.

### Verify all four client modes

> Use the Tazo skill to check my setup, then verify sync upload, async upload,
> sync streaming, and async streaming with the bundled weather recording.
> Check that the selected choice is `get_weather` and save a JSON report locally.

The bundled `check_setup.py` helper checks the SDK installation, key presence,
health, and model discovery without submitting audio. Its `--offline` option
checks only local prerequisites.

The `verify.py` helper uses a sample saying "What is the weather today?".
Pass `--models tazo-large` to run all four modes against the available API model.
A successful run prints `Summary: 4/4 passed`. Tazo Small is unavailable through
the API; contact [support@tazolabs.com](mailto:support@tazolabs.com) for assistance. The report includes
decisions, timings, errors, and SDK metadata; it excludes the key and raw audio.
Your agent can find these helpers relative to the installed skill directory.

> [!NOTE]
>
> Verification submits real inference work and counts toward your account's
> [limits](https://docs.tazolabs.com/guides/limits.md). For a smaller check, ask your agent to use
> `--models tazo-large --modes sync-upload`. The expected weather choice applies
> to the bundled sample; choose an expected answer appropriate to your own audio.

## Supported operations

The skill uses these four public operations through the Python SDK:

| Operation | API |
| - | - |
| Uploaded audio decisions | `POST /v1/decisions` |
| Streaming decisions | WebSocket `/v1/decisions/stream` |
| Model discovery | `GET /v1/models` |
| Health | `GET /healthz` |

Both synchronous and asynchronous clients are supported. Create and manage
accounts and API keys in the [console](https://app.tazolabs.com).

## Update or remove the skill

Run these commands from the project where the skill is installed:

```bash theme={null}
npx skills@latest list
npx skills@latest check
npx skills@latest update tazo
```

To uninstall it:

```bash theme={null}
npx skills@latest remove tazo
```

For global installations, add `--global` when listing, updating, or removing
the skill. See the
[skills CLI reference](https://github.com/vercel-labs/skills/blob/main/docs/commands.md)
for supported agents and installation options, or the
[skill README](https://github.com/tazolabs/agent-skills#readme) for running the
helpers directly.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.