Brisvox vs Pipecat vs LiveKit Agents

Brisvox, Pipecat and LiveKit Agents are open-source voice agent frameworks that differ mainly in their runtime model and in whether a media server is needed. Brisvox runs as one Rust process with WebSocket, Twilio and Telnyx built in. LiveKit Agents needs a LiveKit server, Cloud or self-hosted. Pipecat needs no media server for its WebSocket and SmallWebRTC transports.

Competitor facts checked on against each project's own documentation. Measured rows are from 14 to 17 October 2026. LiveKit Agents was not measured.

Last updated:

What was measured

These rows come from this project's own benchmark harness, with the result file named in the last column. "Not measured" means the repository has no such run for that system. None of this is an independent benchmark: no third-party density comparison of the three was found.

Measured here Brisvox LiveKit Agents Pipecat 1.12.0 Result files
Concurrent calls within the latency SLO (p95 voice-to-voice at most 1110 ms), 4 vCPU / 4 GB. Median, runs in brackets 250 (260, 250, 220) Not measured 1 (1, 1, 0) 4 vCPU, SLO runs
Most concurrent calls with no failed call and no lost turn (any latency), 4 vCPU / 4 GB 430 (430, 480, 400) Not measured 10 (10, 10, 10) 4 vCPU, max runs
Server memory at that maximum, 4 vCPU / 4 GB (PSS summed over the container’s processes) 753 MB at 430 calls Not measured 2.4 GB at 10 calls 4 vCPU, max runs
Concurrent calls within the latency SLO, 8 vCPU / 8 GB 470 (480, 460, 470) Not measured 10 (10, 5, 10) 8 vCPU, SLO runs
Most concurrent calls with no failed call and no lost turn, 8 vCPU / 8 GB 780 (740, 780, 800) Not measured 10 (10, 20, 10) 8 vCPU, max runs
Server memory at that maximum, 8 vCPU / 8 GB 1.3 GB at 780 calls Not measured 3.3 GB at 10 calls 8 vCPU, max runs
Per call at the 4 vCPU / 4 GB maximum: memory and CPU, idle baseline included 1.8 MB, 0.007 cores Not measured 239.9 MB, 0.137 cores 4 vCPU, max runs
Voice-to-voice on real providers, English, p50 / p95 (5 concurrent calls, 100 turns, Deepgram Aura voice; Windows, public internet) 1.85 s / 2.88 s Not measured 1.97 s / 4.35 s Live run
From this project's own benchmark harness. Capacity and memory rows: fake providers with fixed latencies, Linux cloud containers, 16 and 17 October 2026. Latency row: real providers, one Windows machine, 14 October 2026. Method and caveats

What each project documents

These rows are what each project's own documentation and repository said on 14 October 2026. They are not benchmarks. Each competitor cell links the page it was read from.

Documented, Brisvox LiveKit Agents Pipecat
Language Rust engine; your agent in Python or TypeScript (SDKs on PyPI and npm) or in Rust Python (livekit-agents 1.8.7); Node.js in a separate repository livekit-agents on PyPIlivekit/agents-js Python (pipecat-ai 1.12.0) pipecat-ai on PyPIpipecat-ai/pipecat
License MIT or Apache-2.0 Apache-2.0; the turn-detector text model and v1-mini use the LiveKit Model License livekit/agentsLiveKit: turn detector BSD-2-Clause pipecat-ai/pipecat
Runtime model One process, one isolated async session per call, models on a dedicated thread pool One OS process per job, “each in its own process for isolation” LiveKit: job lifecycle Documented as one session per process or container Pipecat: running bots in production
Needs a media server No: WebSocket, Twilio and Telnyx built in; WebRTC and SIP are cargo features Yes: a LiveKit server (Cloud or self-hosted), the agent joins a room livekit/agents No for WebSocket and SmallWebRTC; the Daily and LiveKit transports use those vendors’ infrastructure Pipecat: transports
Telephony Twilio and Telnyx media streams, SIP trunk listener, DTMF, transfer LiveKit SIP service: Twilio, Telnyx, Plivo and others, DTMF, cold and warm transfer LiveKit: telephony Twilio, Telnyx, Plivo, Exotel serializers; no SIP stack of its own found in the pages read (inferred) Pipecat: transports
Turn detection Smart Turn v3.2 (ONNX, on device) with an adaptive policy Own turn-detector models: a text model (396 MB on disk, about 50 to 160 ms per turn on CPU) and the audio model v1-mini (size and cost not stated) LiveKit: turn detector Smart Turn v3 (8 MB int8) Pipecat: Smart Turn overviewpipecat-ai/smart-turn
Turkish First class, with its own quality notes Listed by both turn detectors LiveKit: turn detector Listed by Smart Turn pipecat-ai/smart-turn
Interruptions Turn epochs; the context keeps only the text actually played Adaptive mode separates interruptions from backchannel and resumes after a false one LiveKit: turns Cancels LLM and TTS, flushes audio, commits only the played text; a minimum-words turn-start strategy and a Krisp backchannel filter Pipecat: interruptions
Tools read_only and side_effect kinds, idempotency keys, never speculative; written in Rust, Python or TypeScript Function tools, async tools, handoff, MCP (Python), provider-side tools; parallel calls not stated on the page read LiveKit: tools Parallel calls by default, async tools, per-tool timeouts, cancel on interruption; MCP not checked Pipecat: function calling
Speech-to-speech OpenAI Realtime, Gemini Live Many realtime models LiveKit: realtime models Several pipecat-ai/pipecat
Overload handling Admission control and graceful drain built in Default load function (5 s average CPU, threshold 0.7) LiveKit: deploymentsLiveKit: server options None found in the docs; autoscaling guidance Pipecat: running bots in production
Hosted cloud None, self-host only Optional LiveKit Cloud livekit/agents Optional Pipecat Cloud Pipecat: running bots in productionPipecat Cloud: sizing
Vendor sizing guidance None; see the measured rows 4 cores / 8 GB for 10 to 25 jobs; a vendor test of 30 agents peaked at about 3.8 cores and 2.8 GB LiveKit: deployments Pipecat Cloud profile of 0.5 vCPU / 1 GB per session Pipecat Cloud: sizing
What each project's own documentation and repository said on 14 October 2026. Not benchmarks. "Not stated" and "inferred" are marked as such in the research notes, where every fact carries its URL.

Does Brisvox need a media server?

No. Brisvox serves WebSocket, Twilio and Telnyx directly, and WebRTC and SIP are optional cargo features. LiveKit Agents needs a LiveKit server because the agent joins a room. Pipecat needs no media server for WebSocket and SmallWebRTC, and uses vendor infrastructure for the Daily and LiveKit transports.

Is Brisvox faster than Pipecat or LiveKit Agents?

Against Pipecat 1.12.0, measured here: on the live run with real providers, the median voice-to-voice time was 1.85 s for Brisvox and 1.97 s for Pipecat, a gap smaller than the LLM provider's own variance. The clearer differences were at the tail (p95 2.88 s against 4.35 s) and in capacity with fake providers (250 against 1 concurrent calls within the latency SLO on 4 vCPU / 4 GB).

LiveKit Agents has not been measured, so there is no answer for it here.

What happens when a caller interrupts?

Brisvox numbers each user turn as an epoch and drops stale frames at every stage, and after a barge-in its LLM context holds only the text that was played. Pipecat's documentation says it cancels the LLM and TTS, flushes audio and commits only the played text. LiveKit Agents' adaptive mode separates interruptions from backchannel and resumes after a false interruption.

In 0.1.4 (November 2026), Brisvox also resumes after a false interruption by default: it pauses at a barge-in candidate and, within a short decision window, either interrupts or continues the same response.

How do side-effect tools behave?

Brisvox marks tools as read_only or side_effect. Side-effect tools never run speculatively and never run twice for one call id, and an identical repeat in the same turn is refused. The exact scope, and the idempotency key to add for "once per order, ever", are in the tool-calling contract. An equivalent guarantee was not found in the Pipecat and LiveKit pages read, which is not the same as it being unavailable.

Which one should you choose?

  • Brisvox if you want one self-hosted process with telephony and browser transports built in, density per container that you can re-measure from committed result files, first-class Turkish, and per-call usage reporting. You accept a 0.x project, first released in October 2026, with no hosted service.
  • Pipecat if you want a Python framework with its own client SDKs, its Daily and LiveKit transports, or a hosted option (Pipecat Cloud).
  • LiveKit Agents if you already run a LiveKit server, want a managed cloud (LiveKit Cloud), or want its SIP service, its turn-detector models and agents in Node.js.

This last list is an editorial reading of the tables above, not a measurement.

Sources

Fetched on 14 October 2026. The full notes, with what is inferred and what a page did not state, are in docs/research/frameworks-2026-10-14.md.

Found something out of date? Open an issue and it will be corrected.