Pipecat

Our full take

Pipecat is two things that share a name, and getting them straight is the whole story. The first is the Pipecat framework, an open-source Python project from Daily (the video and audio infrastructure company) that wires together the parts a voice agent needs: speech-to-text, a language model and a voice, plus the transport that carries the call. It has roughly 12,600 stars on GitHub and a BSD-2-Clause licence, which is to say it is genuinely free and you can use it commercially. The second is Pipecat Cloud, Daily’s paid managed hosting that runs your Pipecat agent at scale so you do not have to keep your own servers alive.

Here is the part that trips people up. Pipecat does not give you a voice, a language model or a speech recogniser of its own. It is vendor-neutral on purpose. You choose every component (Deepgram or Whisper for transcription, OpenAI or Anthropic for the brain, ElevenLabs or Cartesia for the voice) and Pipecat is the wiring that makes them talk to each other in real time. That is powerful if you want control, and it means you pay each of those providers directly, separately from anything you pay Pipecat.

So this is a developer framework, not a finished product. There is no drag-and-drop builder and no dashboard where a non-technical buyer assembles an agent. You write Python. If you do not have a developer who is comfortable in Python, this is the wrong starting point, and that is worth knowing before the free price tag pulls you in.

Now the money, because the free framework and the paid cloud are billed completely differently.

Self-hosting the framework costs you nothing in Pipecat fees. You run it on your own server and pay only your chosen providers. For a small project that can genuinely be the cheapest route there is, with the catch that running production voice infrastructure (autoscaling, recovery, telephony plumbing) is real engineering work you now own.

Pipecat Cloud takes that work off your plate and bills per running agent. The base size, agent-1x, is $0.01 a minute while an agent is active, and a tiny $0.0005 a minute when reserved (that is one-twentieth of the active rate, a way to keep instances warm during quiet periods so calls connect instantly). Bigger agents cost more: agent-2x is $0.02 a minute, agent-3x is $0.03. On top of that you choose how the call reaches a phone. A Daily phone number over the normal network (PSTN) is $0.018 a minute, a SIP connection to your own phone supplier is $0.005 a minute, and one-to-one browser voice calls on Daily’s own WebRTC transport come with free voice minutes for one human and one agent.

Put the pieces together for a typical phone agent and the workings look like this: $0.01 hosting + $0.018 telephony + whatever your speech-to-text, language model and voice providers charge. The speech-to-text, language-model and voice lines read zero in our table on purpose, because those are not Pipecat’s to bill. Add a Deepgram transcription cost, an OpenAI token cost and an ElevenLabs voice cost and a realistic all-in lands somewhere around $0.03 to $0.20 a minute, depending heavily on which models you pick. Recording is extra if you want it ($0.005 a minute for audio), and Krisp noise filtering is included up to 10,000 minutes a month.

On compliance, Pipecat Cloud is in a stronger spot than most open-source tooling because Daily sits underneath it. Daily’s documentation states that Pipecat Cloud runs on HIPAA-compliant infrastructure and that Daily can sign a single Business Associate Agreement covering Pipecat Cloud, the Daily WebRTC layer and transcription, with HIPAA controls enabled per account on request. The product page also states the service is GDPR-compliant and that Daily does not store your data. We have marked HIPAA and GDPR true on the strength of those primary pages. SOC 2 we have left unticked: Daily describes it as on the roadmap for Pipecat Cloud specifically rather than as a published certificate we can link today, so we are not claiming it.

My read: Pipecat is one of the best places to start if you want full control of the stack and you have the Python skills to use it. You are not locked into anyone’s voice, model or pricing, the framework is free, and when you outgrow your own servers the managed cloud is cheap and the telephony and HIPAA story are already handled. The honest limit is that none of this is no-code. You assemble the agent, you wire and pay for every provider, and you maintain it. If you want to point and click your way to a working agent on a single bill, a bundled platform will get you there faster.

The 1 to 10 scores on this page are an editorial preview, our provisional read to get the framework in place, not a measured result. We have not run Pipecat through our own call tests yet, so there is no Voxrater latency figure here. The pricing, capability and compliance detail is sourced from the Pipecat homepage and GitHub, plus Daily’s Pipecat Cloud pricing, product and HIPAA pages, captured 2026-05-31.