Legacy CustomTkinter interface
The mature desktop interface that remains the legacy foundation for the Tauri migration.
Evidence ledger: five source-classified artifacts
Evidence that keeps its source and boundary attached.
Five source-classified artifacts describe visible exports, log-derived telemetry and a reported migration record. They are not interchangeable samples.
- 4h 27m 35s
- Measured agenda duration
- Log-derived
- 104 pages
- Visible 1 Aug UI export
- Visible UI artifact
- 285
- Generations: 120 cloud / 165 local
- Log-derived
- 70 min
- Raw visible chat case
- Qualitative n=1
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.
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.
Product demo
Watch the current Kira walkthrough
See the current agenda, chat-context, co-host reaction, and interface flow before reviewing the dated build progress.
Audio: Spanish · Captions: auto-generated Spanish; YouTube may offer automatic English translation
Spanish audioYouTube
Current OpenCohost Kira walkthrough
The walkthrough shows the product direction in motion. Kira remains in active development and this demo is not a release promise.
Open on YouTube
Read the English demo summary
English summary of the walkthrough
This English summary covers agenda control, co-host reactions, chat context, voice coordination, and the current interface direction. The player provides auto-generated Spanish captions; availability and quality of YouTube automatic English translation can vary.
SOURCE-CLASSIFIED EVIDENCE
Original source artifacts with readable context first
Each stratum gives an accessible HTML summary before its source PDF. Visible UI labels, raw visible chat, log-derived values and reported work records answer different questions.
Visible
Visible UI artifact
1 Aug raw long-session export
A 104-page UI export shows a 4h27m operating window at minute resolution. It is not a complete transcript or analytical telemetry.
- Visible
- 188 visible KIRA agenda-output markers; 6 operator messages and 6 visible direct-response headers with queue labels of 10s, 12s, 45s, 55s, 1s and 21s.
- Visible
- 4 cloud-down/restored cycles, 29 scheduled retries plus 4 forced retries, 23 visible guardrail rejections and 10 visible memory saves.
- Caveat
- Markers and headers are what the UI made visible; they do not establish full transcript coverage, causal performance or aggregate telemetry.
Measured
Log-derived
1 Aug telemetry analysis
The retained telemetry report is a separate log-derived analysis, not a reading of the visible UI export.
- Measured
- 4h 27m 35s, 285 generations, 120 cloud and 165 local; local fallback totals 2.62 h (58.6% of the session).
- Measured
- 4 cloud restorations and 194 prefetches are reported from traced runtime events.
- Caveat
- Its exact metrics are bounded to that instrumented run and are not a production guarantee.
Measured
Log-derived
30 Jul traced stress audit
An independent earlier sample: the raw 1 Aug export is newer than the 30 Jul audit, but does not supersede it.
- Measured
- 2h50m25s active agenda/audio, 169 completions, 139 TTS pipelines, and 1,625 fragments across 2h27m32s of audio. First fragment: median .37s, p95 .54s.
- Measured
- 86/87 accepted prefetch gaps were at most 100ms (median 46ms); context mean was 68.6%, with 1,126 pairs evicted across 15 topics.
- Visible difference
- Four direct questions were answered 14 to 29 minutes late in the 30 Jul audit; six 1 Aug visible queue labels were 1 to 55s. This is an observed difference suggesting improvement, not a guarantee.
- Caveat
- GLM5.2 and audible Unicode are operator observations, not log-proven. This audit does not test factual reliability, personality or production readiness.
Reported
Reported engineering record
2 Aug core-domain migration record
A migration report synthesized from a work record, not an independent live-repository audit.
- Reported
- 27 modules moved; 150 obsolete references in 45 files; 21 identical-blob renames plus 6 audited import/docstring changes.
- Reported
- It records 6 entrypoints, compileall, 5,142 collected tests, three introduced defects repaired, and a final reported 5,129 passed / 14 skipped plus one regression.
- Not demonstrated
- It does not show that main contains the migration, that the 7,009-line llm_engine.py God Object is solved, full DDD adoption or a production regression-free state.
Visible
Raw visible chat export
70-minute agenda chat export
A raw visible chat export records one 70-minute agenda conversation. It is a qualitative n=1 case, not telemetry or an evaluation study.
- Visible
- 70 minutes from 14:12 to 15:22; 10 topics; 59 Kira interventions and 2 operator interventions.
- Caveat
- This is a raw visible chat export; subjective usefulness is self-report, not productivity or learning proof.
EVIDENCE BOUNDARY
Not demonstrated
- Production-ready streaming, validation with external streamers, or viewer retention.
- Reliable live-chat understanding in real communities or factual reliability in sensitive domains.
- Safe autonomous hosting without human supervision.
- Productivity or learning gains from the single self-report chat case.
- A complete transcript, universal performance guarantee or independent audit of the reported migration.
CTK ABANDONED / TAURI SOURCE OF TRUTH
From abandoned prototype to source of truth
Since mid-July 2026 the Tauri app is the source of truth: the improved, stable version of the product. CustomTkinter (CTK) was a functional prototype, fully abandoned (Legacy Unstable) without the interruption improvements.
- Agenda continuity
- CTK · ABANDONED PROTOTYPE
Prototype evidence without interruption improvements - TAURI · SOURCE OF TRUTH
Source of truth: improved and stable - Stream reactions
- CTK · ABANDONED PROTOTYPE
Functional but irregular prototype behavior - TAURI · SOURCE OF TRUTH
Source of truth: current development - Memory and chat
- CTK · ABANDONED PROTOTYPE
Functional prototype behavior - TAURI · SOURCE OF TRUTH
Source of truth: improved and stable - User experience
- CTK · ABANDONED PROTOTYPE
Abandoned prototype - TAURI · SOURCE OF TRUTH
The current product interface
DIRECTION, NOT DATES
Project direction, without artificial dates
The horizon labels describe intent, not delivery promises. Circularity and factual discipline remain open research questions until they have suitable evidence.
- Now
Refine the Tauri source of truth and preserve observable failure states.
- Next
Prove stream reactions in Tauri and collect structured external feedback.
- Later
Broaden reproducible model and hardware matrices without universal claims.
- Open research
Circularity, factual discipline, conversational novelty and safer live behavior.