Legacy CustomTkinter interface
The mature desktop interface that remains the legacy foundation for the Tauri migration.
OpenCohost Kira — Active development
OpenCohost Kira is an AI co-hostyou can run locally
The open beta runs on your own machine with Ollama, the models you choose, and your own hardware.
The open beta is available locally. Run OpenCohost locally with your own Ollama, models, and hardware; there is no hosted service or bundled model.
Real product progressSee the current Tauri buildFewer silent stretches
Kira can suggest a transition, question, or reaction when the room goes quiet.
Chat becomes context
She summarizes common intent instead of reading spam line by line.
Streamer stays in control
Agenda, voice input, timing, and guardrails remain supervised by you.
Live runtime capture
Use an AI as a companion, a teacher, or to kill time
An uncut runtime capture of the real product interface: turn-based conversation, its own voice, and failure recovery. Not a mockup.
Audio: Spanish · English capture in progress
What works today
A supervised co-host loop, not an autonomy promise.
Kira can carry a prepared agenda, take direct input and recover across local and cloud providers. The streamer remains the editor and factual authority.
Product- 01Verified
Agenda continuity, Direct/PTT, chat input, BYOK and Ollama paths.
- 02Experimental
Stream connection and reactions were functional but uneven in the abandoned CTK prototype; Tauri carries them forward.
- 03Legacy Unstable
The CustomTkinter (CTK) prototype was fully abandoned in mid-July 2026; it lacks the interruption improvements.
Measured session
Long-form continuity with the limits left visible.
A dated instrumented run shows sustained agenda flow and provider recovery. It does not prove production streaming or autonomous hosting.
- 4 h 27 min
- Exact duration 4:27:35
- agenda session
- 285
- Traced generations
- runtime total
- 120 / 165
- Cloud / local generations
- provider split
Streamer outcomes
Useful when the show needs continuity, not another dashboard.
- 01
Keep a prepared segment moving
Agenda turns create a recoverable line of continuity while the host can interrupt or redirect.
- 02
Choose where inference runs
Use tested local model paths or bring a cloud key without pretending the trade-offs disappear.
- 03
Inspect what failed
Runtime traces, status labels and dated reports make limitations part of the product story.
Dated product progress
Source of truth since mid-July 2026
The current Kira interface
Since mid-July 2026 the Tauri app is the source of truth: the improved, stable version of the product. The CustomTkinter (CTK) build was a functional prototype, fully abandoned (Legacy Unstable) without the interruption improvements.
Current and historical build evidence
Each image is paired by a stable identifier so assets and bilingual descriptions stay aligned.
Tauri migration in progress
A current migration snapshot; the transition is still underway.
Bring-your-own-key provider setup
Current setup work for user-provided cloud credentials without promising provider availability.
Cloud model exploration
Evaluation of possible cloud model options; this is not an availability promise.
Current co-host panel direction
The newer panel direction for agenda, reactions, and co-host status.
Compact logs and status
A compact view of live chat, text-to-speech, and status signals.
Agenda keeps moving
A concept view of Kira carrying stream topics while the host is away.
Fewer dead-air moments
A concept view of Kira helping maintain momentum during a quiet stretch.
Local-first hardware matrix
Is your setup ready for Kira?
Local mode avoids cloud/API billing, but the real load is handled by your hardware: GPU, VRAM, and games.
| GPU | Game Type | Experience |
|---|---|---|
| RTX 3060 12GB | Light / Competitive (Fortnite, Valorant, LoL) | ✅ Lightweight voice mode, responsive and fluid co-hosting.Tested |
| RTX 3060 12GB | Heavy / Ultra 3D (Cyberpunk, Alan Wake 2) | ⚠️ Lightweight voice mode recommended. Minor delay possible.Estimate |
| RTX 4090 | Any game / settings | ✅ Best headroom for heavier local modes; still depends on model, game load, and active features.Estimate |
| GTX 1660 | Lightweight games (Fortnite, Valorant) | ✅ Lightweight voice mode; disabling high-poly avatar recommended.Estimate |
| No Dedicated GPU | Just chatting screen / browsing | ❌ Local voice processing needs a dedicated GPU.Estimate |
OpenCohost Kira is a solo, self-funded personal project, so the hardware lab is literally one PC: an RTX 3060 with 12 GB of VRAM. That single rig is the only "Tested" row here. Every other GPU is an educated estimate, not a benchmark. Real latency and performance depend on your exact GPU, your free VRAM, and your active game load.
Self-hosted beta
Download the open beta for Windows.
OpenCohost is a local-first open-source application and local AI co-host. The alpha installer is generated by CI and published on GitHub Releases. There is no hosted service and no bundled model: you bring Ollama, choose and download the models, and provide the hardware.
v0.2.0-alpha.1 · 97.7 MB · Windows 10/11 x64 · published Aug 21, 2026 · Verify the SHA256 checksums
Alpha build, generated by CI. If something misbehaves, report it — that feedback is what the beta is for.
Local beta feedback
Use GitHub Issues to improve the self-hosted beta.
Report reproducible setup, workflow, memory, or integration problems without API keys, private logs, or sensitive information.
FAQ · 03
Three honest answers before you evaluate Kira
What is Kira today?
A beta streaming co-host with supervised agenda, Direct/PTT, chat, memory and local/cloud model paths. It is not an autonomous host.
Can I use local or cloud models?
Both paths exist. Ollama has been tested with specific Gemma, Llama and Qwen configurations; BYOK supports cloud providers. Results depend on the exact model and hardware.
Is streaming production-ready?
No. The Tauri app is the source of truth, but streaming has not been validated with external streamers or viewers; the CTK prototype was fully abandoned.