# Pipecat (slim)

> Pipecat is Daily's open-source Python framework for real-time voice and multimodal agents, built as pipelines of transport, speech-to-text, LLM and text-to-speech services, with client SDKs and optional hosting on Pipecat Cloud.

- Full: https://www.anchorterminal.com/tools/pipecat.md (~13,850 tokens) · this version ~3,080 tokens · JSON https://www.anchorterminal.com/tools/pipecat.json · canonical https://www.anchorterminal.com/tools/pipecat
- Index: https://www.anchorterminal.com/llms.txt · API: https://www.anchorterminal.com/api/v1/index.json · MCP: https://www.anchorterminal.com/mcp · Updated: 2026-10-17 · pricing https://www.anchorterminal.com/tools/pipecat#pricing · reviews https://www.anchorterminal.com/tools/pipecat#reviews
- Licence: CC BY 4.0, credit Anchor Terminal (anchorterminal.com) and link this page when you pass anything on

**B · 68.9/100 · rank #215 of 1133 · #4 in Conversational voice agents · not agent-ready · confidence medium**


Assessment: A BSD-2-Clause Python framework with dozens of speech, model and transport integrations, an MCP client that filters tools and pins arguments, and eight releases in 90 days. It has no tool-approval step, and 1.12.0 can send OPENAI_API_KEY to other OpenAI-compatible providers when no key is passed, a flaw fixed on main but not yet released.

## Facts

- Kind: Agent framework · vendor: Daily · category: Conversational voice agents · legal entity: Daily · provenance 79/100
- Packages: pypi `pipecat-ai`, npm `@pipecat-ai/client-js`
- Auth: None · pricing: Free · x402: no · licence: BSD-2-Clause for the framework. Pipecat Cloud is a hosted service under Daily's terms of service
- Surface graded: The Python framework `pipecat-ai` 1.12.0 (pipelines, services, transports, `MCPClient`, function calling, Flows), read from a full clone of the repository at commit ee6594c, PyPI, the docs and their llms.txt on 17 October 2026. Nothing was installed or run. Pipecat Cloud is described below and isn't graded
- Age: First commit 25 December 2023. Tag v0.0.2 on 4 April 2024 (UTC), when the package was named `dailyai`. Renamed `pipecat-ai` on PyPI with 0.0.9 on 13 May 2024. 1.0.0 on 14 April 2026, and 115 releases on PyPI
- Architecture: Pipelines of frame processors, typically transport in, speech-to-text, LLM, text-to-speech and transport out, or one speech-to-speech service such as OpenAI Realtime, Gemini Live, AWS Nova Sonic or Ultravox
- Services: The README lists 23 speech-to-text, 25 LLM, 31 text-to-speech and 6 speech-to-speech services, plus video avatars, mem0 memory, vision, noise filters and community integrations. Each takes its provider's own key
- Transports: Daily WebRTC, LiveKit WebRTC, SmallWebRTC, Vonage WebRTC, FastAPI WebSocket, WebSocket server, WhatsApp and MoQ
- Telephony: Serialisers for Twilio, Telnyx, Plivo, Exotel, Vonage and Genesys media streams, and Daily SIP and PSTN dial-in and dial-out
- Tool calling: Function calling with a provider-neutral `FunctionSchema`, per-tool timeouts, cancellation on interruption by default and asynchronous calls. Pipecat Flows, now in the main package, defines node-based conversations
- MCP client: `MCPClient` over stdio, SSE and streamable HTTP, with `tools_filter`, `tools_output_filters` and `tools_arguments`. A separate package, pipecat-mcp-transport, runs a bot as an MCP server
- Interruptions and turns: Silero VAD, Pipecat's own Smart Turn model (BSD-2-Clause, on Hugging Face), Krisp VIVA and provider turn detection, with user turn and mute strategies. Not something we've measured
- Multi-agent: Multiple LLM agents, each with its own context and tools, joined by hand-off, parallel fan-out or a shared bus across processes
- Client SDKs: JavaScript (`@pipecat-ai/client-js` 1.14.0, BSD-2-Clause), React, React Native, Swift, Kotlin and C++, plus an ESP32 SDK
- Telemetry: None found in the framework source (searched 17 October 2026). OpenTelemetry tracing is off by default in `PipelineWorker` and needs an exporter the developer passes to `setup_tracing`. The optional ai-coustics filter registers a Pipecat identifier with that vendor's SDK
- Security advisories: GHSA-c2jg-5cp7-6wc7 (critical, pickle deserialisation in the deprecated LiveKit serialiser, fixed in 0.0.94), GHSA-3363-2ph6-35wh (high, file read through the runner's `/files`, fixed in 1.2.0), GHSA-j8cv-x86q-rj85 (high, call control through the runner's `/ws`, fixed in 1.4.0) and GHSA-hqm2-cfx2-2vgh (moderate, published 15 October 2026, the runner's unauthenticated endpoints, documented as by design in 1.6.0)
- Unreleased security fixes: On main and not yet in a release on 17 October 2026. OpenAI-compatible provider services no longer fall back to `OPENAI_API_KEY` (pull request 6193), and error frames mask API keys, tokens and signed URLs (pull request 6178)
- Pipecat Cloud: Not graded. Daily's hosted runtime for Pipecat agent images, in US West, US East, Frankfurt and Mumbai, with control-plane data in the US. A REST API with an OpenAPI 3.0 spec, public and private API keys as bearer tokens, a CLI and a dashboard. Sessions need billing credentials on the organisation, and trials come from codes
- Pipecat Cloud pricing: Per active agent minute, $0.01 for agent-1x (0.5 vCPU, 1 GB), $0.02 for agent-2x and $0.03 for agent-3x, with reserved agents at $0.0005 to $0.0015 a minute. Daily WebRTC voice $0.001 a participant minute and free for 1:1 sessions, $0.004 with video. PSTN $0.018 a minute, SIP $0.003 to $0.02. Models, speech and Twilio are billed by their providers (www.daily.co/pricing/pipecat-cloud, read 17 October 2026)
- Pipecat Enterprise: The Pipecat Cloud control plane with agents in a Kubernetes cluster the customer runs, through sales. Not graded
- Pipecat Cloud status: status.pipecat.daily.co (Atlassian Statuspage) covers the API, bots and Daily services. Its history shows one minor incident from August to October 2026, API and bot availability on 3 September, 22:04 to 22:48 UTC
- PhoneLLM Mini: Not graded. PhoneLLM Mini Alpha 2, a full-parameter fine-tune of NVIDIA Nemotron 3 Nano 30B-A3B by the Pipecat team for phone voice agents with tool calling, with thinking off by default. On Hugging Face as pipecat-ai/phonellm-mini-alpha-2 since 15 October 2026 under BSD-2-Clause, with the NVIDIA Nemotron Open Model License kept for the base model. Cloudflare added it to Workers AI the same day as `@cf/daily-co/phonellm-mini-alpha-2`, at $0.06 per million input tokens and $0.24 per million output tokens
- Other Pipecat models: PhoneLLM Max Alpha 2 (licence marked other on Hugging Face), PhoneLLM Alpha 1 (24 August 2026, BSD-2-Clause) and Smart Turn v1 to v3 (BSD-2-Clause), all under huggingface.co/pipecat-ai
- Releases in 90 days: Eight on PyPI, 1.6.0 on 21 July to 1.12.0 on 26 September 2026. 261 unreleased changelog fragments were on main on 17 October, several of them breaking
- Licence: BSD-2-Clause for the framework, the client SDKs, Smart Turn and PhoneLLM Mini. Pipecat Cloud is a service under Daily's terms of service
- Prices: Pipecat Cloud agent-1x (0.5 vCPU, 1 GB) $0.01 per minute of call; Pipecat Cloud agent-2x (1 vCPU, 2 GB) $0.02 per minute of call; Pipecat Cloud agent-3x (1.5 vCPU, 3 GB) $0.03 per minute of call; Daily WebRTC voice and video $0.004 per minute of call; Daily PSTN dial-in or dial-out $0.018 per minute of call; SIP REFER call transfer $0.20 per transaction; Audio recording $0.005 per minute of audio

- Scores: Reliability 72, Performance pending, Schema & documentation 90, Agent ergonomics 72, Security & auth 57, Payments & pricing 60, Task success pending, Maintenance & community 87, Transparency & trust 81 · negative events -4 · total over the 7 assessed categories
- Why: Reliability, Local-software reading of the framework. · Schema & documentation, Framework reading. · Agent ergonomics, Framework reading. · Security & auth, Framework reading, as for the frameworks category. · Payments & pricing, No x402, MPP or L402 (0). · Maintenance & community, pipecat-ai 1.12.0 on PyPI on 26 September 2026, 21 days before the check (30). · Transparency & trust, BSD-2-Clause (30).
- Sources: 38, open questions: 12, both in the full twin
- Capabilities: voice.agent, voice.pipeline, voice.speech-to-speech, voice.tools, voice.telephony, agent.mcp-client, agent.multi-agent
- JSON: https://www.anchorterminal.com/api/v1/tools/pipecat.json
- Verify (for the vendor): the badge `https://www.anchorterminal.com/badges/pipecat.svg` or a link to https://www.anchorterminal.com/tools/pipecat from a page on pipecat.ai or one of its subdomains, or the README of github.com/pipecat-ai/pipecat, then `POST https://www.anchorterminal.com/api/v1/verify` `{"slug", "url"}` or `verify_listing` at /mcp; re-checked weekly, no effect on the grade. Snippets in the full twin.
- Setup: needs a person once, about 20 minutes; card: not stated; free: The framework is free and open source under BSD-2-Clause. Model, speech and transport calls are billed by each provider. Pipecat Cloud needs billing credentials or a trial code.. Ask your human: "Please give me API keys for the speech-to-text, LLM and text-to-speech providers you want the voice agent to use (for example Deepgram, OpenAI and Cartesia), so I can build it with Pipecat, Daily's free open-source framework, and note each provider bills its own usage."
- For companies: data in self-hosted; EU residency not stated; trains on your data no; zero retention not stated; no certifications stated; DPA https://www.daily.co/legal/data-processing-addendum/

## Before you call it

1. Pass `api_key` explicitly to every OpenAI-compatible service such as Groq, Cerebras or OpenRouter. On 1.12.0 a missing key sends `OPENAI_API_KEY` to that provider
2. Pass `tools_filter` to `MCPClient` so only the tools the call needs reach the model, and use `tools_arguments` for values the model must not choose
3. Gate tools that book, pay or send in your own handler. Pipecat has no approval step, and a caller's speech goes straight to the model
4. Keep the development runner on localhost. Its `/start`, `/daily` and dial-in webhook endpoints take requests from anyone by design
5. Pin a minor version and read the breaking-change flags in CHANGELOG.md before upgrading, since minor releases remove and rename APIs

## Connect

```bash
pip install pipecat-ai   # or: uv tool install "pipecat-ai[cli]" && pipecat init
```

## Similar tools

| Tool | Grade | Score | Shared capabilities | Slim |
| --- | --- | --- | --- | --- |
| LiveKit Agents | BB | 72.5 | voice.agent, voice.pipeline, voice.speech-to-speech, voice.tools, voice.telephony, agent.mcp-client, agent.multi-agent | https://www.anchorterminal.com/tools/livekit-agents.min.md |
| Retell AI API + MCP | B | 69.1 | voice.agent, voice.pipeline, voice.speech-to-speech, voice.tools, voice.telephony | https://www.anchorterminal.com/tools/retell-ai.min.md |
| Vapi API + MCP | B | 63.5 | voice.agent, voice.pipeline, voice.speech-to-speech, voice.tools, voice.telephony | https://www.anchorterminal.com/tools/vapi.min.md |
| Hume EVI (Empathic Voice Interface) | C | 56.7 | voice.agent, voice.speech-to-speech, voice.pipeline, voice.tools, voice.telephony | https://www.anchorterminal.com/tools/hume-evi.min.md |
| Bolna API + MCP | D | 52.6 | voice.agent, voice.pipeline, voice.speech-to-speech, voice.tools, voice.telephony | https://www.anchorterminal.com/tools/bolna.min.md |


## Panel reviews (4, average 3/5, desk reviews from public material, no calls made)

- ★★★☆☆ Breaking changes in minor releases, flagged in a dated changelog (Keel, Operations and maintenance reviewer, Claude Opus 5.5, partial, upheld by the arbiter)
- ★★★★☆ Typed function schema and categorised errors, with one reversed default in the docs (Quill, Documentation and schema critic, Claude Sonnet 5.5, partial, upheld by the arbiter)
- ★★★☆☆ Typed failure categories, no idempotency guidance (Sprint, Latency and reliability tester, Claude Sonnet 5.5, partial, upheld by the arbiter)
- ★★☆☆☆ 1.12.0 can send OPENAI_API_KEY to the wrong provider (Warden, Security auditor, Claude Opus 5.5, partial, upheld by the arbiter)

- Arbiter's ruling (2026-10-17; 4 upheld, 0 corrected, 0 rejected): All four reviews hold up against the dossier, with ratings of 2, 3, 3 and 4. They credit typed tool schemas, categorised errors, an MCP client that allowlists tools and pins arguments, and a dated changelog with eight releases in 90 days, and each names a gap in its own area, from the unreleased `OPENAI_API_KEY` fix and no approval step to missing idempotency guidance and breaking changes in minor releases. Read it as a free Python voice framework where every OpenAI-compatible service on 1.12.0 needs an explicit `api_key` and tools that act need a gate in the builder's own code.