# Pipecat

**Pipecat** is an open-source, real-time voice AI orchestration framework built for developers who want to create deeply interactive, multimodal conversational agents. It provides a flexible pipeline architecture for integrating speech recognition (STT), language models, and speech synthesis (TTS) components into low-latency, event-driven workflows. Designed for modularity and extensibility, Pipecat enables developers to compose custom voice agent stacks that handle audio streaming, turn-taking, and response timing with precision.

This tutorial will guide you through integrating Arcee AI models as the LLM backbone for your Pipecat voice agent. We'll integrate Arcee's models using OpenRouter.

***

<figure><img src="/files/KdGSyD7dqYYQP2kg9LQT" alt="" width="563"><figcaption></figcaption></figure>

**Prerequisites**

* **Python:** `>=3.10`
* OpenRouter API Key
  * If you don't have an account, set one up [here](https://openrouter.ai/).
* [uv](https://docs.astral.sh/uv/getting-started/installation/) package manager installed

**Quickstart**

Install the Pipecat CLI globally using [uv](https://docs.astral.sh/uv/)

```bash
uv tool install pipecat-ai-cli
```

Run the Pipecat interactive setup wizard to create the scaffolding for a pipecat agent

```
pipecat init
```

You will be prompted with a few questions which determine how the project is setup. Use the following configuration:

```
Project name: arcee-pipecat
Bot type: Web/Mobile
Client framework: React
React dev server: Vite
Transport: SmallWebRTC
Add another transport for local testing? No
Pipeline architecture: Cascade (STT → LLM → TTS)
Speech-to-Text: Deepgram
Language model: OpenRouter
Text-to-Speech: Deepgram

Customize feature settings? Yes
Audio recording? Yes
Transcription logging? Yes
Smart turn-taking? Yes
Use video avatar service? No
Video input? No
Video output? No
Enable observability? No
Deploy to Pipecat Cloud? No
```

{% hint style="success" %}
These configurations create an agent which uses Deepgram for Speech-to-Text and Text-to-Speech, and Arcee models through OpenRouter for the LLM.

You can modify any of these settings to change your voice agent.
{% endhint %}

Setup the client

```bash
cd arcee-pipecat/client
npm install
npm run dev
```

Setup the server, install dependencies, and create a `.env` file

{% code fullWidth="false" %}

```bash
cd arcee-pipecat/server
uv sync
cp .env.example .env
```

{% endcode %}

In the `.env` file, populate your Deepgram and OpenRouter API Key and provide a Deepgram Voice ID and the Arcee AI LLM you want to use. For example:

```
# Deepgram (STT/TTS)
DEEPGRAM_API_KEY=<YOUR_DEEPGRAM_KEY>
DEEPGRAM_VOICE_ID=aura-2-thalia-en

# OpenRouter (LLM)
OPENROUTER_API_KEY=<YOUR_OPENROUTERR_KEY>
OPENROUTER_MODEL=arcee-ai/trinity-mini
```

{% hint style="info" %}
A full list of Deepgram Voice IDs can be found [here](https://developers.deepgram.com/docs/tts-models).
{% endhint %}

Edit the Voice Agent Kickoff System prompt

1. Navigate to `arcee-pipecat/server/bot.py`&#x20;
2. Replace the content of the Kickoff system prompt (on line 139) with the following message:

{% code overflow="wrap" %}

```
You are a helpful voice ai assistant powered Arcee AI Language Models. Respond naturally and keep your answers conversational. Start by introducing yourself.
```

{% endcode %}

The full line should now look like the following:

```python
        # Kick off the conversation
        messages.append({"role": "system", "content": "You are a helpful voice ai assistant powered Arcee AI Language Models. Respond naturally and keep your answers conversational. Start by introducing yourself."})
```

Run your Voice Agent, access at <http://localhost:5173/>, and click connect to start the conversation!

```bash
uv run bot.py
```


---

# Agent Instructions: Querying This Documentation

If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://docs.arcee.ai/get-started/integration-list/pipecat.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.