llama.cpp already supports /v1/realtime for local speech-to-speech chat with LFM3-Omni
ngxson · x · 2026-11-16
Developer ngxson points out that llama.cpp's server has supported an OpenAI-style /v1/realtime endpoint for a while, enabling fully local speech-to-speech chat with omnimodal models like LFM3-Omni.
Per the README, the API supports both realtime (conversation) and transcription session types, takes PCM16 mono audio input with server-side resampling, and is configurable via session.update — no OpenAI API needed to build a real-time voice assistant locally.
More from coding & agent
- Dev teases open-source computer use for iPhones with 2.1ms typing, no extra hardware — iamrobotbear · 2026-11-16
- Building an autonomous game-clip pipeline with Codex without burning usage limits — gondorzabitasi · 2026-11-16
- AI agent swarm of 20 Claude models proves first nontrivial lower bounds for Naslund's crossing problem, formalized in Lean — burny_tech · 2026-11-16
- ywza: MCP server lets AI control your PC via accessibility tree — Important_Earth6615 · 2026-11-16
- Interactive Airflow 3 ML orchestration tutorial, AI-generated and open-sourced — Total_Firefighter_59 · 2026-11-16
- Codex Users Hit Recurring Permission Errors on Staging Folders, Deploy Debugging Now Outlasts Build Times — Revan_84 · 2026-11-16