LLM through Ollama
Neural workload 1Local reasoning and co-host responses without making cloud/API billing the default path.
Ecosystem chassis
OpenCohost is the chassis that makes the local AI stack usable on stream.
The big achievement is coordinating local LLM behavior, privacy-aware voice output, voice context, chat reaction, agenda control, and failure resilience as one streamer-ready ecosystem.
The achievement
One machine can run or connect multiple AI workloads: Ollama for the local LLM, a TTS path that can stay fully local through Piper, optional Edge-TTS light synthesis when privacy policy allows it, and the LiveAudio transcription path with Silero VAD plus OpenAI Whisper open-source transcription. OpenCohost adds the orchestration, context management, repetition mitigation, and resilience that make this practical. Hardware and VRAM decide how far you can push it.
Inside OpenCohost
Local LLM, TTS, profiles, chat reactions, context, resilience.
OpenCohost chassis
Orchestration, context, agenda, reactions, resilience, streamer controls.
LiveAudio bridge
VAD, transcription, subtitles, voice context.
Inside OpenCohost
Privacy-aware TTS stack
Neural workload 2Voice output can stay local through Piper, use optional Edge-TTS light synthesis when allowed, and evolve toward custom voice work without hiding privacy tradeoffs.
Profiles and personalization
An OpenCohost-designed layer for streamer-approved personality, style, and boundaries.
Chat reaction algorithm
A proprietary reaction layer that turns chat pressure into useful co-host moments.
Context and repetition mitigation
OpenCohost manages context and reduces repeated local-LLM behavior instead of dumping raw prompts at the model.
Agenda Mode
Semi-automatic topics with human approval, designed for control, flow, and failure resilience.
LiveAudio bridge
Silero VAD
Detects real speaking moments so voice interaction does not treat silence as work.
OpenAI Whisper open-source transcription
Neural workload 3Turns speech into text for subtitles, transcripts, and usable voice context.
LiveAudio simplification layer
LiveAudio is a separate connected app that packages VAD, transcription, subtitles, and voice context.
Optional ambience
Avatar module
Visual state for the audience. Useful, but not required for the core intelligence loop.
Music module
Ambient control and ducking for broadcast feel. Optional, not part of the required AI chassis.