Add fully local voice agent example (examples/voice_agents/local_only)#6525
Open
thefullnacho wants to merge 2 commits into
Open
Add fully local voice agent example (examples/voice_agents/local_only)#6525thefullnacho wants to merge 2 commits into
thefullnacho wants to merge 2 commits into
Conversation
|
|
Composes existing pieces into a zero-cloud voice pipeline: speaches (faster-whisper) STT and kokoro-fastapi TTS via the openai plugin's base_url, Ollama LLM via with_ollama(), silero VAD, and local turn detection. Includes a per-stage latency benchmark and a README with measured numbers on consumer hardware. See livekit#6522 for the proposal discussion. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Qoh9oPdHqMXMt58gAwxhgK
thefullnacho
force-pushed
the
local-only-example
branch
from
July 23, 2026 09:56
5848fb0 to
543069b
Compare
![devin-ai-integration[bot]](https://avatars.githubusercontent.com/in/811515?s=60&v=4){kind=small}
This comment was marked as resolved.
This comment was marked as resolved.
Sorry, something went wrong.
The setup step still said qwen3:4b from before the default switched to llama3.2:3b. A user following the quickstart verbatim would pull the model the failure-modes section warns against, then hit model-not-found. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Qoh9oPdHqMXMt58gAwxhgK
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Proposed in #6522. A voice agent where every stage runs on your own hardware, zero cloud calls:
openai.LLM.with_ollama()→ OllamaIncludes
bench.py, a per-stage latency benchmark, and a README with measured numbers (RTX 5080 + RTX 4060 Ti, warm path, medians of 5): TTS first byte 36ms, STT 359ms for a ~3s utterance, LLM TTFT 127ms at 170 tok/s. Roughly 520ms from end of speech to first agent audio, from consumer hardware.The README also documents the failure modes I hit building it, because the next person will hit them too: thinking models stall the voice path (qwen3 streamed its first content token at 18s, and none of Ollama's switches fix it cleanly), prebuilt GPU images without sm_120 kernels, and cold-start vs warm-path benchmarking.
One open question from the issue stands: the turn-detector plugin is deprecated in favor of
inference.TurnDetector, which routes through the inference gateway. This example uses the plugin with a comment explaining why, happy to update when there's a local-capable replacement.🤖 Generated with Claude Code
https://claude.ai/code/session_01Qoh9oPdHqMXMt58gAwxhgK