pipedog

[!IMPORTANT] Unofficial. pipedog is an independent project. It is not affiliated with or endorsed by Daily or the Pipecat project. The name is a nod to Pipecat; the code is a port of it, under the same BSD 2-Clause license.

bunx @pipedog/cli init       # or: npx @pipedog/cli init

You build a bot as a pipeline of frame processors, and frames flow through it:

The API mirrors Pipecat's: the same classes, frames, events, module paths and wire protocols, with camelCase members. So Pipecat's client SDKs (@pipecat-ai/client-js, @pipecat-ai/small-webrtc-transport, voice-ui-kit) work against a pipedog bot unchanged.

Sponsors

Top sponsor: Nexi Health. A medical device AI startup that runs pipedog in production, handling 100+ calls every day.

Why TypeScript?

• One language across the stack. Bot, backend and web client in TypeScript: shared types, one toolchain, no separate Python service to deploy and operate.
• Runs where your app runs. Bun, Node, Deno and Cloudflare Workers, bundled to a single file. Deploy a voice agent to serverless or the edge, close to your users.
• Less overhead. About 5× the frame throughput of Pipecat on CPython, 10–14× lower per-frame latency, and cold starts around 45 ms instead of 400 ms (Performance).
• The JavaScript ecosystem. npm packages, your existing deploy pipeline, and the language most web teams already write.
• Still Pipecat. Same architecture, frames and wire protocols, so Pipecat's client SDKs, docs and patterns carry over.

Quickstart

A voice assistant you talk to in the browser, with OpenAI speech-to-text, chat and text-to-speech over SmallWebRTC.

The quickest way to a project like this is the CLI, which writes the bot, its package.json, a README and a guide for your coding agent for the services you pick (docs):

bunx @pipedog/cli init       # or: npx @pipedog/cli init

Or write it by hand:

Install

bun add @pipedog/core @pipedog/asyncio @pipedog/openai @pipedog/transport-smallwebrtc \
  @pipedog/audio-silero @pipedog/audio-smart-turn onnxruntime-node

(npm install, pnpm add and deno add npm:... work the same way.) onnxruntime-node runs the local Silero VAD and Smart Turn models; without it, [email protected] (WASM) is used instead.

Write the bot

bot.ts:

以上為 README 的開頭部分。其餘內容請閱讀原文。 檢視完整 README ↗