Pipecat Cloud pricing (2026): agent hosting rates, what's passed through, and a 10,000-minute example

Two things share the Pipecat name. Pipecat is the open-source framework — BSD-2-Clause, "Maintained by Daily and the community" (GitHub, accessed 2026-09-08) — and it costs nothing to use. Pipecat Cloud is the paid managed service that runs agents built with it: "managed infrastructure for deploying and scaling Pipecat voice agents to production, run by the Pipecat team" (Pipecat Cloud introduction, accessed 2026-09-08). This guide covers the second one. Its pricing page lives on Daily's site (pipecat.ai/pricing returned a 404 on review); every number below comes from that page and the docs that define the billing unit.

Pipecat Cloud publishes rates for the infrastructure layer and passes the AI layer through. For LLM, STT, and TTS the page says "Bring your own provider" and "Billed to provider" (Pipecat Cloud pricing, accessed 2026-09-08). The published numbers are a floor, not a total.

How Pipecat Cloud's pricing model works

Pipecat Cloud "uses usage-based pricing" (introduction, accessed 2026-09-08). There are no subscription tiers, only per-minute line items plus a "Contact sales" path for "Volume Pricing" (Pipecat Cloud pricing, accessed 2026-09-08).

The core line item is agent hosting, priced per minute by instance profile. The pricing page lists the rates and the deploy docs define the sizes (pricing; deployment docs, both accessed 2026-09-08):

• agent-1x — 0.5 vCPU, 1 GB, the default, "Best for voice agents": $0.01/min active, $0.0005/min reserved.
• agent-2x — 1 vCPU, 2 GB: $0.02/min active, $0.0010/min reserved.
• agent-3x — 1.5 vCPU, 3 GB: $0.03/min active, $0.0015/min reserved.

The scaling docs define the two categories. "An active session starts when you call the /start endpoint (or CLI or SDK equivalent) and ends when your agent's pipeline shuts down." Reserved minutes are "Time your warm agent instances are kept running, even when idle." "Both active and reserved session time is measured to the second and billed in minutes." (Pipecat Cloud scaling docs, accessed 2026-09-08.)

Two consequences. "Each instance runs one active session at a time (one bot per instance)," so a larger profile buys one session more CPU and memory, not more concurrent calls. And warm capacity bills continuously: "A single instance kept warm with --min-agents 1 accrues 1,440 reserved minutes a day, even with zero active sessions." The minimum defaults to 0, with a "free auto-scaling buffer" above your reserved instances (same page).

What's included vs. billed separately

Source: Pipecat Cloud pricing, accessed 2026-09-08.

The page mentions "80+ service providers" to connect and an "Enterprise Integrated Billing" option — "bundled billing for inference at enterprise scale" — with no published rate. Its cost calculator takes your providers' inference rates as inputs; they are not Pipecat Cloud prices.

Worked example: 10,000 minutes a month

Take a common workload: 2,000 calls × 5 minutes = 10,000 active session minutes per month, over PSTN, one agent-1x instance per call, no warm instances, no transfers. All arithmetic is illustrative, from displayed prices.

Agent hosting (agent-1x): 10,000 × $0.01 = $100

PSTN dial-in/dial-out: 10,000 × $0.018 = $180

Published infrastructure subtotal: $280/month

Optional published add-ons on the same workload:

• Audio recording: 10,000 × $0.005 = $50
• Krisp VIVA: 10,000 minutes sits exactly at the free allowance; minute 10,001 onward bills $0.0015
• One warm instance all month: 1,440 × 30 = 43,200 minutes × $0.0005 = $21.60
• agent-2x instead of agent-1x: hosting becomes $200; agent-3x: $300

What the example cannot include is the rest of the bill. Speech-to-text, the language model, and text-to-speech are "Billed to provider," at whatever rates and models you choose. They are usually the largest share of a voice agent's per-minute cost, and none appears on Pipecat Cloud's page as a number. Phone-number rental is not itemized either. $280 is the floor of the infrastructure layer for this workload, not the price of the agent.

Gotcha checklist

• The hosting rate is not the agent's cost. Add your STT, LLM, and TTS provider rates per minute before comparing Pipecat Cloud to anything with an all-in price.
• Reserved minutes run around the clock. --min-agents 1 bills 1,440 reserved minutes a day whether or not anyone calls — about $21.60 a month per warm agent-1x instance, more on larger profiles.
• The active clock runs from /start to pipeline shutdown, not from answer to hang-up. Setup and teardown seconds are billable.
• One session per instance. A larger profile does not raise concurrency; it costs 2× or 3× per minute for the same single session.
• Telephony is a separate line. PSTN adds $0.018/minute — more than the agent-1x hosting rate itself — and SIP REFER transfers are $0.20 per event.
• Noise suppression turns paid at 10,000 active minutes. Above the allowance, Krisp VIVA bills $0.0015/minute.

How ThunderPhone prices differ

This is a comparison between layers. Pipecat Cloud hosts agent code you write, and your inference providers bill the AI. ThunderPhone is a finished phone agent: the pipeline already exists, and it publishes one all-in per-minute engine rate — speech recognition, language model, and voice included — pay as you go with no subscription, seat fee, or platform fee: Spark 2¢/minute, Bolt 5¢/minute, Storm 9¢/minute (unpacked in ThunderPhone pricing explained). The same 10,000-minute scenario computes directly: $200, $500, or $900 per month by engine (illustrative; a premium voice or language path adds up to 3¢/minute, hold bills a flat 2¢/minute, and telephony depends on setup).

The honest flip side: if you want to write the agent — choose each speech, model, and voice vendor, own the turn-taking, add custom processors — Pipecat is the right choice, and Pipecat Cloud is a clean way to run it without operating servers. Depending on provider picks, a Pipecat Cloud agent can land below or above ThunderPhone's rates. The difference is that one total is computable before you deploy and the other is discovered as you select providers. ThunderPhone is GDPR and HIPAA compliant — its published pricing lists no compliance add-on fee — and customers remain responsible for configuring their workflows, data handling, recordings, and notices for their own obligations.

FAQ

How much does Pipecat Cloud cost per minute?

$0.01, $0.02, or $0.03 per active minute for hosting by instance size, plus telephony and add-ons — with the AI models billed separately by your providers. Reserved (warm, idle) instances bill $0.0005–$0.0015 per minute. Pipecat Cloud pricing, accessed 2026-09-08.

Is Pipecat free?

The Pipecat framework is free and open source (BSD-2-Clause); Pipecat Cloud is a paid, usage-based hosting service. Its page lists no free hosting tier, though WebRTC voice transport is free for 1:1 sessions and Krisp VIVA is free up to 10,000 active minutes a month. GitHub and Pipecat Cloud pricing, accessed 2026-09-08.

Does Pipecat Cloud's price include speech-to-text, the LLM, and text-to-speech?

No. Those lines are marked "Bring your own provider" and "Billed to provider"; an "Enterprise Integrated Billing" option is mentioned without a published rate. Pipecat Cloud pricing, accessed 2026-09-08.

What would 10,000 minutes a month cost on Pipecat Cloud?

About $280 in published infrastructure charges on agent-1x with PSTN ($100 hosting + $180 PSTN), before inference. Recording adds $50 and a single warm instance about $21.60. STT, LLM, and TTS are not computable from Pipecat Cloud's page. For contrast, ThunderPhone's all-in rates put the same workload at $200/$500/$900 by engine, illustratively.

Sources and freshness

All Pipecat and Pipecat Cloud claims were checked on September 8, 2026 against the Pipecat Cloud pricing page, the introduction, scaling, and deployment docs, and the Pipecat GitHub repository; ThunderPhone pricing reflects its published pricing as of September 1, 2026. This guide is documentation-based — no deployment was built and no invoice reviewed. Two figures deserve care: the recording storage rate is shown per minute without a stated counting rule, and the SIP rate is a range ($0.003–$0.02). Confirm against the live pages before relying on them.