pipecat

pipecat

Pipecat is a leading voice agent in the 2026 AI agent engineering landscape (13.1k★).

When to use this skill

Frame/processor/pipeline architecture for realtime voice + multimodal agents
40+ service plugins: STT/TTS/LLM providers, transports (Daily WebRTC, WebSocket, telephony)
Pipecat Flows for structured conversation state machines
Client SDKs: JS, React, iOS, Android + voice UI kit
Vendor-neutral: swap any provider per pipeline stage
Used by NVIDIA ACE and Daily production stacks
Ecosystem overlap: Pairs with livekit-agents as the two OSS voice lanes

Installation

uv add pipecat-ai   # extras: "pipecat-ai[option,...]"; pip also works

Core API

pipeline = Pipeline([
    transport.input(),
    stt,            # e.g. Deepgram
    context_aggregator.user(),
    llm,            # e.g. OpenAI
    tts,            # e.g. Cartesia
    transport.output(),
    context_aggregator.assistant(),
])
task = PipelineTask(pipeline)
await PipelineRunner().run(task)

Instructions

Step 1: Initialize the environment

Verify you have Python installed and set up properly. Use the correct package manager (pip, npm, pnpm, uv) as specified in the installation section.

Step 2: Establish the agent flow

Architecture pattern: Dataflow pipeline: frames (audio/text/LLM messages) stream through processors; ultra-low-latency scheduling; transports abstract the realtime channel

Step 3: Implement tool integrations and policies

Integrate necessary tools, environment keys, and safety gates according to your architectural needs. Position in the ecosystem: Composable pipeline lane of voice agents vs LiveKit's infrastructure lane; strongest provider neutrality

Examples

Example 1: Minimal Agent Run

pipeline = Pipeline([
    transport.input(),
    stt,            # e.g. Deepgram
    context_aggregator.user(),
    llm,            # e.g. OpenAI
    tts,            # e.g. Cartesia
    transport.output(),
    context_aggregator.assistant(),
])
task = PipelineTask(pipeline)
await PipelineRunner().run(task)

Best practices

Follow the primary installation commands for Pipecat to ensure dependencies are resolved correctly.
Configure model providers and API keys securely before starting the runtime.
Use appropriate sandboxing or permission isolation if executing AI-generated code.

About this skill