A local AI co-host stack for streaming.
OpenCohost coordinates Ollama models, voice output, and streamer-controlled Agenda Mode. LiveAudio is a separate connected app for voice input. The self-hosted alpha has no hosted service or bundled models.
Quick start
Choose the Windows alpha installer or the source-build path, then start Ollama and download a model. Local setup covers both paths; the co-host guide shows the human-approved Agenda Mode flow.
Local requirements
LLM, TTS, and Whisper/VAD workloads compete with OBS and game load. Model choice, GPU/VRAM, RAM, electricity, and setup time affect what is practical; there is no universal hardware profile.
Capabilities
Inside OpenCohost
LLM through Ollama
Neural workload 1Runs the model you choose locally; cloud/API billing is not the default path.
Privacy-aware TTS stack
Neural workload 2Piper can stay local; optional Edge-TTS is available when privacy policy allows it.
Profiles and personalization
Streamer-approved personality, style, and boundaries.
Chat reaction algorithm
A proprietary reaction layer turns chat pressure into co-host moments.
Context and repetition mitigation
Manages context to reduce repeated local-LLM behavior.
Agenda Mode
Semi-automatic topics require human approval.
LiveAudio bridge
Silero VAD
Detects speech activity for voice interaction.
OpenAI Whisper open-source transcription
Neural workload 3Transcribes speech for subtitles, transcripts, and voice context.
LiveAudio simplification layer
A separate connected app for VAD, transcription, subtitles, and voice context.
Optional ambience
Avatar module
Visual state for the audience; not required for the core AI workflow.
Music module
Ambient control and ducking; not required for the core AI workflow.