Development & infrastructureDaily.co (open source community project)
Pipecat
Decision matrix
Pipecat app decision matrix
Overview
Pipecat is a free, BSD-2-Clause licensed framework (maintained by Daily) for building realtime voice and multimodal AI pipelines: compose a pipeline, connect a transport (WebRTC, SIP, PSTN), and wire in any of 200+ integrated speech, language and vision providers. Pick it when you want to assemble a custom voice pipeline from best-of-breed components rather than use one vendor's bundled solution. You integrate and self-host the pieces yourself.
Other good fits
Projects needing telephony-grade transport (SIP, PSTN) alongside WebRTC for voice connectivity.
+ Pros
Read pros
Pipecat is an open-source framework for building realtime voice and multimodal AI pipelines: you compose a pipeline, connect a transport (WebRTC, SIP or PSTN for telephony), and add whichever of 200+ integrated providers you want for speech, language and vision (examples include OpenAI, Soniox and Cartesia). Its pipeline architecture processes data at the frame level with components working concurrently rather than strictly sequentially, and it handles production concerns like interruption handling, turn detection and context management that you'd otherwise have to build yourself.
Pick it when…
You want to mix specific providers (a particular speech-to-text engine, a particular model, a particular TTS voice) rather than accept one vendor's bundled choices.
You need telephony transport (SIP, PSTN) alongside WebRTC, not just browser-based voice.
You want interruption handling and turn detection handled by the framework instead of building that logic from scratch.
− Cons
Read cons
Assembling a pipeline from multiple providers means multiple accounts, multiple bills, and multiple points of failure to monitor, compared to one vendor's bundled API. You're responsible for hosting and integrating the pieces yourself; Pipecat gives you the framework and the connectors, not a managed running service. More provider choice also means more version and compatibility management as each integrated provider evolves independently.
Pick something else when…
You want one managed, bundled API without assembling components yourself.OpenAI Realtime gives you speech-to-speech in one hosted service.
You specifically need a genuinely open, self-hosted full-duplex model as one component.Moshi could be a piece you wire into a Pipecat pipeline, or used directly if you don't need the broader framework.
Your voice needs are simple enough that mixing providers isn't worth the integration effort. A single managed API is less to build and maintain.
Avoid if
You don't need to mix and match providers; a bundled solution from one vendor is simpler if that flexibility isn't worth the setup.
Workflow
Choose your transport (WebRTC for browser, SIP/PSTN for telephony) → pick providers for speech-to-text, the language model, and text-to-speech from Pipecat's integrations → compose the pipeline connecting them → add tool calls where needed → test with a real conversation, specifically including interruptions → tune turn-detection and context settings based on what you observe.
Cost
Pipecat itself is free and open source. Real costs come from two places: the providers you connect (speech-to-text, language model, TTS each bill separately per their own pricing) and your own hosting, since you self-host the pipeline itself. This is meaningfully more setup and integration work than a single managed API, in exchange for choosing exactly which providers power each stage.
Restrictions
You integrate and self-host the pipeline components yourself; Pipecat provides the framework and connectors, not managed hosting. Each of the 200+ integrated providers has its own account, pricing and terms that Pipecat doesn't change or simplify.
First project
Documentation-based suggestion — not run here.
Read first project
Build a minimal voice pipeline and test interruption handling.
Set up a WebRTC transport and pick one speech-to-text provider, one language model, and one TTS provider from Pipecat's integrations.
Compose the pipeline connecting all three.
Have a short test conversation, deliberately interrupting the agent mid-response at least once.
Check whether the interruption was handled naturally and whether turn detection correctly identified when you started and stopped speaking.
Done when you've had one complete conversation with at least one successful interruption, and you can describe specifically how the framework handled it versus how you'd have to build that yourself.
Links & evidence
Documentation-based · not hands-on · Sep 23, 2026
Rests on Pipecat's homepage and documentation site, checked 23 September 2026. No pipeline was built or run for this review.
Assemble voice activity detection, transcription, language, and speech components.
Connect your chosen services or local models through many integrations.
Carry audio over WebRTC or another transport without tying the pipeline to one vendor.
Daily.co (open source community project) · free · self-host
voice activity detection
speech to text
text to speech
local models
webrtc transport
open source pipeline
component integration
open-source
build-app
automate
View sourced imageStill frame from the official Pipecat Tail monitoring demonstration. Tail is a companion terminal dashboard for Pipecat sessions.
Pipecat official product documentation
Owner-requested editorial use of public official interface screenshots, 15 September 2026. Copyright retained by the original publisher; attribution preserved. No open licence or hands-on course execution is asserted. Source asset: https://raw.githubusercontent.com/pipecat-ai/tail/refs/heads/main/tail-image.gif. PNG conversion preserves the complete frame. Converted on 15 October 2026 to a WebP (quality 80) for faster page loads; the original image remains in the media library.