Original voice greeting
A real owner-approved WAV is recorded and available for visitors to hear.
Evelyn Hart / Development & Evidence
Measured progress, clear limits
A dated account of engineering progress, source reviews and what remains untested. The published record distinguishes source checks, simulated software checks, real visual captures and future release requirements.

Development status / October 2026
The current development reports describe Evelyn as a separate Qwen-based AI and technology historian in progress. A historical Kira conversation test harness has been located for adaptation, but Evelyn's original voice greeting has now been recorded and approved. Her public installer, spontaneous model-backed dialogue using this voice, synchronized lip movement, visitor-memory persistence and repeated live-conversation tests remain unverified. A separate controlled eyelid blink has passed.
October 19 development update: Visitor-specific application-memory namespaces and reusable detailed head controls are being added. A frozen 14-turn test with three synthetic visitors is prepared to check delayed recall, restart persistence, visitor separation, unknown visitors and a corrected preference. The test is not yet executed. Live recall, audio, geometry and clean installation remain unverified; saved application memory is not proof of neural learning. Documented limits ↗
October 20 head-preview milestone: A newer local compatibility-render test now shows Evelyn’s licensed 3D head candidate from distinct front, three-quarter, and side angles. Developer screenshots report that all three visible saved views passed blank-screen checks and matched their direct canvas captures. The preview uses modified diffuse materials and reduced textures; the original PBR appearance remains unqualified. The controlled eyelid test is now verified; speaking, synchronized lip movement, generated conversation, visitor memory after restart, and a public installer remain unverified. The earlier failed-render evidence remains archived. Read the October 20 milestone → Browse reports and source images →
See Evelyn among Kira Labs projects →Original voice · October 24, 2026 · Founder-approved recording
The founder has approved Evelyn Hart's original voice and confirmed Codex is using it. The 145,964-byte, 24 kHz mono PCM16 WAV is preserved in KiraWorld and is now publicly available here for listening. This is a genuine recording, not a synthetic-input test fixture or an earlier silent preflight.
Listen to Evelyn's approved introduction
“Hello, I'm Evelyn Hart. And I'm glad to meet you.”
Recording: 3.04 seconds · PCM16 mono · 24 kHz. A preserved copy also appears in the private KiraWorld source history. The public website file is the same Git blob as the owner's approved recording.
A real owner-approved WAV is recorded and available for visitors to hear.
Evelyn is the last planned Kira Labs synthetic individual to use the earlier voice generator. Future synthetic people are intended to transition to newer voice technology; Video Studio still uses a version of the older generator.
Arbitrary new utterances, consistent restart playback, unscripted conversation and synchronized facial animation need separately reviewed recordings.
The earlier recorder-wrapper preflight failed and produced no voice. That result remains in the dated history below; this later accepted greeting supersedes the former overall “no approved original voice” status. The public app installer and new replacement voice system are not yet released. KiraWorld voice commit (may require repository access) ↗
Original voice · October 24, 2026 · Attempt 1193 · Recorded failure
The latest committed voice-audition status records a narrowly scoped preflight and one failed attempt. The Qwen3-TTS worker did not start; No approved voice was generated by that failed attempt. A separate, later original voice greeting has since been approved and is available above.
All 13 cached VoiceDesign model files, approximately 4.52 GB total, passed complete current file hashing; interpreter and output-path metadata also passed preparation.
The recorder detected unexpected additional process generations and rejected the launch. Cleanup returned both owned process Jobs to empty and closed all 16 retained known handles, but complete identity accounting was not established.
No Evelyn WAV, audible playback, synchronized lip movement or action footage came from this recorded attempt. A silent baseline or narrated text preview would not substitute for an actual recording.
Next qualified step: review the unexpected recorder process lineage with a smaller no-model startup diagnostic, without weakening identity or cleanup guards. Later attempts visible in ongoing local development must be verified separately before the website reports a changed outcome.
Release status: a separate unsigned Voice Designer 0.1.6 installer preview now has tested backend installation, reinstall and local voice generation on one Windows host; native GUI controls and clean-PC release qualification remain unverified. Evelyn's own companion installer remains unreleased. Read draft computer requirements and troubleshooting → · October 24 source record (private repository access may be required) ↗
Evelyn Hart and Avatar Builder · October 23, 2026 · Independently reviewed source
Three new engineering steps passed their described independent and primary source-code reviews. None supplies new live conversation audio, a naturally moving full body, or an installer ready for public download.
A checked design now maps 25 joint chains to a canonical 163-bone rig and 139 skin-weight groups. The export auditor compares the package, build report, vertex reference, GLB, Blend and rig-rest files. It verifies declared joint/hierarchy, rest-transform and sampled skin data—not natural deformation, full anatomy, realistic clothes or physical motion. The new export execution gates remain disabled.
The app now has reviewed source fixes for preserving a selected library destination and a supplied document title/date while recursive discovery is in progress. The staged ordinary-file library remains 30 publications plus two original AI/film-technology primers. These are files in preparation, not material Evelyn has actually read. Installation, import, page-reading and restart behavior still need runtime tests.
A new source-reviewed adapter retains Evelyn's original adult female voice design, 55-word audition text, profile and seed. The planned test records before generation and plays the generated voice through admitted physical speakers while retaining receipts. The staged app's 28-file mapping was reviewed. No unique voice has yet been generated, played, recorded or approved.
Not yet established: human-audible speech, live visitor conversations, full internal anatomy, accepted natural movement, fabric motion, installer operation, or an exported body copied and verified in Patient Horizon. Failed and partial future recordings remain part of the planned review. Private books, reference media and visitor memories are not published.
Read the October 23 engineering journal → · Preserved library assembly →
Original research documents · October 20, 2026
The three pictures above are actual compatibility-render screenshots. The source report records front, three-quarter and side views accepted by the saved-evidence checks. This establishes static visibility only; the result is not photorealistic or a speaking avatar.
Read report 70: verified head captures and test method → Read report 65: 52 historian reference cards →
Earlier correction preparation (before the subsequent passing test): A source-reviewed change searched for sufficient eyelid movement while checking eye stability and return to rest. At that earlier snapshot the new run had not begun due to available RAM below the 8 GiB guard. A separate later blink/reopening test passed the fixed criterion; generated voice remains unverified. Read the new technical update →
Separate motion test — did not pass. Both eyelids moved only about 1.7% against the predeclared 65% blink-reduction requirement. Jaw calibration was accepted, but six integrated motion captures and the video were not produced. Software audio-clock tests do not verify actual speech. Read the motion failure record →
There is still no public installer, validated generated conversation, synchronized talking head, or verified memory persistence after restart.
Evelyn Hart · October 23, 2026 · Verified preparation / pending release tests
The latest repository review records a two-page release-requirements document and a four-page installation and troubleshooting guide. All six PDF pages were visually inspected, and the prepared PDF, text and HTML materials were checked against copies in the university review folder. This is a documentation milestone, not a public installer, published end-user guide or confirmation that Evelyn can run on another computer.
Read the public previews: Evelyn release requirements and acceptance checklist → · Installation and troubleshooting topics →. Both are labeled as development plans, not verified user instructions.
The documents explain planned signed updates, protected starting identity, developing memories, owner-controlled data, an editable local library, and practical help for installation, documents, voice and recovery. Before public distribution, the guide must be matched to the tested installer and bundled offline as well as placed beside the official download.
Reviewed body-motion source now combines rest-relative movements with blended poses and carries breathing timing between movement routes. Static review passed; full-body export parity, recorded natural walking, reaching, clothing movement and unseen body variants have not passed runtime testing.
The original-voice execution bridge has a reviewed change to retain real process handles before potentially failing setup steps. Evelyn's voice will be selected from a newly generated original adult female design suitable for her young-adult character. No original audio, qualified physical playback, recorded conversation or final voice approval exists yet.
Release boundary: signed installation, tested hardware requirements, backup/restore, book discovery, privacy behavior, original voice, natural movement and third-party clean installation still need validation. The full four-page guide remains a pre-release draft. A public troubleshooting topic preview and a separate release-requirements roadmap are available, but neither is a tested installer manual or a download. October 23 document review →