Pipecat alternatives: open-source and hosted voice agents
Updated By Agent Radar6 min read
Short answer
Pipecat alternatives compared: LiveKit Agents, OpenAI Agents SDK, TEN Framework, Vapi and Retell AI, with licenses, per-minute prices and lock-in notes.
Top pick: LiveKit Agents. Apache-2.0 framework where the agent joins a LiveKit room, with SIP telephony and Node.js support.
Quick comparison
| Pick | License | Pricing | Source |
|---|---|---|---|
| 1.LiveKit Agents | Apache-2.0 | Free framework; self-host the LiveKit server or pay for LiveKit Cloud. | livekit/agents on GitHub |
| 2.OpenAI Agents SDK (voice and realtime) | MIT | Free SDK; you pay OpenAI API usage. | openai/openai-agents-python on GitHub |
| 3.TEN Framework | Apache-2.0 with additional conditions | Free framework; the default examples use Agora RTC credentials. | TEN-framework/ten-framework on GitHub |
| 4.Vapi | Not stated | $0.05/min hosting plus pass-through STT, LLM, TTS and telephony; $5 free credits. | Vapi pricing |
| 5.Retell AI | Not stated | $0.07-$0.31/min pay-as-you-go; 20 concurrent calls included. | Retell AI pricing |
Most people looking for a Pipecat alternative want one of two things: a framework that bundles the media layer instead of leaving the transport to them, or a hosted platform so they do not run voice infrastructure at all. The first group should look at LiveKit Agents and the OpenAI Agents SDK; the second at Vapi and Retell AI.
Short answer
If you want open source with the media server and telephony in one project, use LiveKit Agents. If you build on OpenAI models and want voice in the same SDK as your text agents, use the OpenAI Agents SDK. If you want a hosted platform and per-minute billing, try Vapi when you want to choose providers, or Retell AI when the job is phone calls. TEN Framework is worth a look for its own turn detection, but read its license first.
Pick by what you are building
| You want | Start with | Watch out for |
|---|---|---|
| Open source with media and SIP included | LiveKit Agents | Tied to LiveKit rooms |
| Voice on OpenAI models, few moving parts | OpenAI Agents SDK | Limited transports |
| Graph-based pipelines with built-in turn detection | TEN Framework | License limits, Agora defaults |
| Hosted, choose your own providers | Vapi | $0.05/min plus provider costs |
| Hosted phone agents | Retell AI | $0.07-$0.31/min |
The order is editorial, not a latency benchmark.
What you give up by leaving Pipecat
Pipecat is BSD-2-Clause and builds a bot as a pipeline of frame processors, with the transport as one processor among others. That is why the same bot can run over Daily, plain WebSockets, a LiveKit room or a Twilio, Telnyx, Plivo, Exotel, Genesys or Vonage media stream. Version 1.16.0, released on November 12, 2026, added a jambonz serializer and diarization speakers across most STT services. If transport freedom and a wide provider list are why you picked it, every alternative below narrows one of them.
The alternatives
LiveKit Agents
LiveKit Agents is Apache-2.0, with about 15,200 GitHub stars on November 12, 2026 (Pipecat has about 17,200). The agent joins a LiveKit room as a server-side participant, and LiveKit's SIP stack handles phone calls into the same rooms as web users. It has semantic turn detection and a Node.js SDK; Pipecat servers are Python only. The cost is that your media runs through a LiveKit server, self-hosted or on LiveKit Cloud. Our Pipecat vs LiveKit Agents comparison covers turn-taking, telephony and testing side by side.
OpenAI Agents SDK
The OpenAI Agents SDK is MIT-licensed in Python and JavaScript. The Python package adds voice pipelines that chain speech-to-text, an agent workflow and text-to-speech (openai-agents[voice]), plus realtime agents on OpenAI's realtime models. The JavaScript SDK's RealtimeAgent runs low-latency voice agents in the browser. You keep tools, handoffs and guardrails from text agents, but you get far fewer transports and phone integrations than Pipecat. See OpenAI Agents alternatives for the wider picture.
TEN Framework
TEN Framework, backed by Agora, builds voice agents as graphs of extensions and ships its own VAD and turn detection models. Its examples include a voice assistant over RTC or WebSocket, and the quick start asks for an Agora App ID and certificate. The license is Apache-2.0 with additional conditions: you may not host it on end-user devices such as phones, or deploy it in a way that competes with Agora's offerings. That rules it out for some products.
Vapi
Vapi is a hosted platform. Its pricing page lists a $0.05 per minute hosting fee, with transcription, model, voice and telephony billed at the provider's price on top, and $5 of free credits to start. Vapi telephony, WebSockets and Daily WebRTC are listed as free transports, while Twilio, Vonage and Telnyx minutes are passed through. It suits teams who want Pipecat-style provider choice without running servers.
Retell AI
Retell AI is a hosted platform focused on phone agents. Its pay-as-you-go plan lists $0.07 to $0.31 per minute for voice agents depending on the models and voices you pick, with 20 concurrent calls included. You trade control over the pipeline for not having to operate it.
How to switch without losing a week
Take one Pipecat bot with real call recordings or test scripts. Rebuild it on the alternative with the same STT, LLM and TTS providers, then compare time to first audio, how often it talks over the user, and cost per minute. For hosted platforms, price your real monthly minutes before you port anything else.
Sources checked November 12, 2026: release notes, official repositories, GitHub license metadata and the Vapi and Retell AI pricing pages. We did not benchmark latency. For neighbouring guides, see Pipecat vs LiveKit Agents, best free AI voice models and the voice agents category.
The picks
Apache-2.0 framework where the agent joins a LiveKit room, with SIP telephony and Node.js support.
Pros
- Media server, job dispatch, telephony and agent framework from one project.
- Python and Node.js SDKs.
Cons
- Built around LiveKit rooms, so the transport is not a free choice.
- Running the media server yourself is real infrastructure work.
Pricing Free framework; self-host the LiveKit server or pay for LiveKit Cloud.
MIT-licensed SDK with realtime agents and STT-agent-TTS voice pipelines, smoothest on OpenAI models.
Pros
- Same agents, tools, handoffs and guardrails as text agents.
- Realtime agents run in the browser with the JavaScript SDK.
Cons
- Few transport and telephony options compared with Pipecat.
- Realtime agents are tied to OpenAI's realtime models.
Pricing Free SDK; you pay OpenAI API usage.
- Rank 3
TEN Framework
Apache-2.0 with additional conditionsOpen sourceGraph-based framework for real-time voice agents from Agora, with its own VAD and turn detection models.
Pros
- Ships its own VAD and turn detection components.
- Examples cover RTC and WebSocket voice assistants.
Cons
- The license bars hosting on end-user devices and uses that compete with Agora.
- Examples lean on Agora App IDs for media.
Pricing Free framework; the default examples use Agora RTC credentials.
- Rank 4
Vapi
Hosted voice agent platform that charges a $0.05 per minute hosting fee plus model and telephony costs at provider price.
Pros
- Phone numbers, WebSockets and WebRTC without running servers.
- Pick your own STT, LLM and voice providers.
Cons
- Closed platform; agent logic lives in its configuration.
- Per-minute fees add up at call-center volume.
Pricing $0.05/min hosting plus pass-through STT, LLM, TTS and telephony; $5 free credits.
- Rank 5
Retell AI
Hosted voice agent platform for phone calls, priced at $0.07 to $0.31 per minute pay-as-you-go.
Pros
- Built for phone agents, with telephony included.
- No infrastructure to run.
Cons
- Closed platform with less control over the pipeline.
- Price depends on the model and voice mix you choose.
Pricing $0.07-$0.31/min pay-as-you-go; 20 concurrent calls included.
Questions
What is the best open-source Pipecat alternative?
LiveKit Agents, if you are happy to run on LiveKit rooms. It is Apache-2.0, has Python and Node.js SDKs and includes SIP telephony. The OpenAI Agents SDK (MIT) is the lighter choice if you build on OpenAI models. Our [Pipecat vs LiveKit Agents](/blog/pipecat-vs-livekit-agents) guide goes deeper.
Pipecat vs Vapi: which should I pick?
Pick Pipecat when you want the code, the transport and the providers under your control and can run the servers. Pick Vapi when you want phone and web voice agents without infrastructure and accept $0.05 per minute on top of model and telephony costs.
Is TEN Framework fully open source?
Its license is Apache-2.0 with extra conditions: you may not host it on end-user devices such as phones, or deploy it in a way that competes with Agora's offerings. Read the LICENSE file before building a product on it.