AI-meeting v0.2.0-beta.2 — voice orb and validation refresh

AI Meeting beta.2 keeps the first run small: speak a plan, confirm the proposed change, and find the saved task next time.

Try the account-free task flow · Try the avatar/audio preview · Watch the 45-second task demo · Project website

If the workflow is useful, star the repository to keep it in your list and follow future releases. For implementation or customization, use the private business inquiry form.

What changed

• The voice orb acknowledges clear microphone input during the VAD handoff, while speaking, thinking, mute and terminal states keep their priority.
• Added regression coverage and documented the input hint and privacy boundary.
• Refreshed the English and Japanese quickstarts, Codespaces entry point and validation evidence.

Download and run

The oss-web.zip is a prebuilt avatar/audio demo. Extract it, run python3 -m http.server 8000 in the extracted directory, and open http://localhost:8000/vrm-demo.html. No API key is needed for this preview.

The source.zip is the source at this release tag. Use Node.js 22, Corepack and pnpm 10.12.2; follow the README to connect an AI provider. SHA-256 checksums are attached. Application-specific source is Apache-2.0; model, SDK and service terms remain separate.

Validation and scope

On September 14, 2026, 1,100 automated tests across 124 files passed in CI, including type checking and the OSS distribution audit. The recorded task demo uses synthetic Japanese input with real Gemini responses in a dedicated recording layout. It is a reproducible demonstration, not a claim about human adoption or long-running meeting quality. The public hosted voice demo has separate provider limits; Meet/Zoom connectors remain experimental.

日本語

音声で決めたことを確認して保存し、次回へ引き継ぐベータ2です。音声入力の最初のフレームでもオーブが入力中を示すようにし、返答中・推論中・ミュート・終了状態の優先順位は維持しています。登録不要のタスク体験、APIキー不要のVRM・音声プレビュー、45秒の操作デモを公開しています。自動テスト1,100件（124ファイル）がCIで合格しました。実人間の利用実績やMeet/Zoomの本番品質を示すものではありません。

AI-meeting v0.2.0-beta.1 — voice, avatars, and tasks

AI Meeting is available as an Apache-2.0 application beta: realtime voice conversation, VRM avatars, interruption, and session-local voice-controlled tasks.

Try the account-free task flow · Try the avatar/audio preview · Watch the 45-second task demo · See the 8-second official TikTok demo · Project website

What this beta includes

• Browser-local VRM audio playback with no API key, plus self-hosted AI provider adapters.
• Immediate playback interruption and confirmation for proposed task changes.
• Task deduplication across whitespace and full-width transcription variants; malformed status values are rejected atomically.
• English/Japanese documentation, recording captions and transcript, and focused quickstarts for the supported avatar paths.

Download and run

The oss-web.zip is a prebuilt avatar/audio demo. Extract it, run python3 -m http.server 8000 in the extracted directory, and open http://localhost:8000/vrm-demo.html. Python 3 is required for this local server example. Opening the HTML through file:// is not supported.

The source.zip is the source at this release tag. Use Node.js 22, Corepack and pinned pnpm 10.12.2; follow the README to build or connect an AI provider. Both archives include applicable license notices. SHA-256 checksums are attached.

Validation and scope

On September 13, 2026, 1,099 automated tests across 124 files passed, along with workspace type checking, production build, and the OSS distribution audit. GitHub CI verifies subsequent published commits.

The video contains real Gemini responses and saved task changes, with synthetic Japanese input and scripted confirmation in a dedicated recording layout. It is a recorded demonstration, not a claim about human adoption or long-running meeting quality. The public hosted voice demo has separate provider limits; the account-free task flow stays in this browser. Meet/Zoom connectors remain experimental. See measured results and known limits.

Application-specific source is Apache-2.0. The pixiv VRM sample, optional integrations, model weights and services retain their own terms.

日本語

音声でタスクを追加し、割り込み、変更案を確認するベータ版です。音声認識による空白・全角半角の揺れで同じタスクが重複する問題を修正しました。

まずは登録不要のタスク体験またはAPIキー不要のVRM・音声プレビューから試せます。45秒の操作デモと公式TikTokの8秒実演も公開しています。AI会話には別途プロバイダー設定が必要です。

ローカルVRMプレビュー、任意のLiveAvatar、Tavusには方式ごとのクイックスタートがあります。必要条件、プライバシー境界、第三者サービスの条件を記載しています。

有料の導入・カスタマイズ・運用支援は非公開の相談フォームからご相談ください。テスター募集ではなく、登録だけで使える月額サービスでもありません。公開の質問や非機密の利用例はGitHub Discussionsへお願いします。

If the project is useful, a GitHub star helps you find it again later and makes it easier for others to discover.