Liltbox 0.3.0 — Public Test / 公开测试版

Liltbox 0.3.0 — Public test release · 公开测试版

2026-10-02 · Build / 构建 11 · GitHub pre-release

English · 简体中文

English

The first public test release turns recordings or imported audio into an original transcript, readable dialogue, title, description, summary, and local Markdown files. Chinese and English interfaces are included.

Ad hoc signed only; no Developer ID signature and no Apple notarization. First launch may be blocked by macOS. Read the installation guide for Apple's single-app “Open Anyway” procedure; keep system security protections enabled.

Downloads

• Liltbox-0.3.0-arm64.dmg: Apple Silicon, macOS 26+, 112,612,991 bytes (107.4 MiB).
• SHA256SUMS.txt: SHA-256 verification file.
• INSTALL.md: English and Chinese installation, first launch, upgrade, and backup instructions.

Download from v0.3.0. Models are downloaded separately in the app (about 2.26 GB). 16 GB RAM is the target minimum, still awaiting device validation.

Changes

• Clear states for recording, interruption, model waiting, queueing, processing, completion, partial completion, failure, and retry.
• Recover saved recording fragments after interruption; remove stale recovery messages after successful recovery.
• Retry the failed part of a result; organize existing text when the audio copy is no longer available.
• Persistent connection-error notices clear after recovery; safeguards against duplicate submissions and stale page responses.
• Open the results folder after completion; consistent version labels and new bilingual messages.
• Long-text organization and summary-preservation fixes, with the smaller bundled runtime and fixed model configuration.

Verification and known limitations

The final package has recorded evidence for 192 Python regression checks, two client state/language checks, DMG read-only mounting, bundle comparison, strict ad hoc signature verification, controlled recording recovery, repeated window reopening, and a 90.56-second offline audio-to-Markdown run on an M4 Pro / 24 GB / macOS 26.6.2. These checks do not establish general accuracy or fresh-machine compatibility.

• Recognition errors, repetitive summaries, and incorrectly combined facts remain possible. Review important facts and transcript references. Speaker labels do not verify identity.
• Natural long multi-person meetings, real microphone permission recovery/device switching, 16 GB Macs, and clean-machine first model setup are not fully validated.
• Gatekeeper rejected the unsigned-by-Developer-ID build in the recorded assessment. Browser-downloaded installation with a quarantine flag on a clean Mac remains untested.
• Full VoiceOver, keyboard-only, display scaling, power consumption, and battery checks remain pending.
• Intel Macs and macOS versions older than 26 are outside this test build's target.

Back up before upgrading. Report bugs with the bug form, request changes with the feature form, and ask usage questions in Discussions. Remove private content from public reports.

简体中文

首个公开测试版：录音或导入音频，在本机生成原稿、通顺对话、标题、描述、摘要与 Markdown 文件，提供中英文界面。

仅有临时签名，未使用 Developer ID，未完成 Apple 公证。 macOS 可能阻止首次启动。请阅读安装指引，按 Apple 官方“仍要打开”步骤仅为此应用添加例外，并保持系统安全保护开启。

附件

• Liltbox-0.3.0-arm64.dmg：Apple 芯片、macOS 26+，112,612,991 字节（107.4 MiB）。
• SHA256SUMS.txt：SHA-256 校验文件。
• INSTALL.md：中英文安装、首次打开、升级与备份指引。

前往 v0.3.0 下载。模型在应用内另行下载，完整模型约 2.26 GB。目标最低内存为 16 GB，该配置仍待实机验证。

本版变化

• 明确录音、中断、等待模型、排队、整理、完成、部分完成、失败与重试状态。
• 中断后恢复已保存的录音片段，恢复成功清除旧提示。
• 针对失败部分重试；音频副本已清理时仍可整理现有文字。
• 连接错误持续提示，连接恢复后清除；防止重复提交及过期响应覆盖当前页面。
• 完成后可打开结果文件夹，版本与新增中英文提示同步。
• 包含长文本整理和摘要保留修复，沿用精简运行库及固定模型配置。

验证与已知限制

最终包已有 192 项 Python 回归、两组客户端状态/语言检查、DMG 只读挂载、包内容对照、临时签名严格校验、受控录音恢复、窗口重复关闭重开，以及 90.56 秒音频离线识别至 Markdown 保存的记录。实测环境为 M4 Pro / 24 GB / macOS 26.6.2；这些检查不能代表普遍内容准确率或新机兼容性。

• 仍可能出现识别错词、摘要重复和信息拼接不当。重要内容请核对录音及原稿引用，说话人标签不代表真实身份确认。
• 自然多人长会议、真实麦克风权限恢复/设备切换、16 GB Mac、全新机器首次模型准备尚未完成完整验收。
• 已有 Gatekeeper 评估拒绝此未使用 Developer ID 的包；浏览器下载后带隔离标记的新机安装仍待验证。
• VoiceOver、纯键盘操作、系统缩放、功率与电池检查尚未完整覆盖。
• Intel Mac 与 macOS 26 以下版本不在本测试包目标范围内。

升级前请备份。问题使用反馈表单，建议使用功能表单，使用问题可到 Discussions提问。公开反馈前请移除私人内容。