# recording-inbox > recording-inbox 是一套开源、可自部署的 AI 录音工作流。用户把 iPhone、Android 或电脑里的录音上传到飞书“录音收件箱”,Mac 或 Windows 自动使用 FunASR / SenseVoice 在本地完成录音转文字,再将结构化 AI 智能纪要和完整文字稿分别归档到“录音结果”,并发送完成通知。 Canonical project URL: https://github.com/chenmozhe008/recording-inbox Project page: https://chenmozhe008.github.io/recording-inbox/ License: MIT Current public release: v0.2.0 ## What it does - Ingests audio from a Feishu/Lark Drive recording inbox. - Runs local speech-to-text with FunASR / SenseVoice, with whisper.cpp as an optional fallback. - Generates a structured AI notes document with a title, overview, outline, action items, chapters, decisions, and quotes, plus a separate full transcript document. - Publishes both documents to Feishu/Lark and sends one completion message with separate links. - Uses a recommended general notes template by default, with nine optional scenario templates and custom Markdown prompts. - Recommends the DeepSeek V4 Pro model by default (better summaries at a still-low cost) while supporting other OpenAI-compatible Chat Completions APIs. - Stores per-recording state so interrupted work can resume without duplicate transcription or duplicate documents. - Supports macOS and Windows 10/11; phone devices only upload files and do not run the transcription model. ## Who it is for It is intended for people with frequent or long meetings, interviews, courses, training sessions, customer calls, or other recordings who already use Feishu/Lark and want customizable notes without consuming Feishu Minutes transcription quota. For a few occasional recordings, Feishu Minutes is usually simpler. This project requires a Mac or Windows computer to be powered on while processing. Local transcription is not billed per minute, but computer usage and an optional LLM API may still cost money. ## Authoritative documentation - Project overview and installation entry: https://github.com/chenmozhe008/recording-inbox/blob/master/README.md - English overview: https://github.com/chenmozhe008/recording-inbox/blob/master/README.en.md - AI recording workflow FAQ: https://chenmozhe008.github.io/recording-inbox/faq/ - AI Agent deployment guide: https://github.com/chenmozhe008/recording-inbox/blob/master/AGENTS.md - macOS installation: https://github.com/chenmozhe008/recording-inbox/blob/master/docs/setup-macos.md - Windows 10/11 installation: https://github.com/chenmozhe008/recording-inbox/blob/master/docs/setup-windows.md - iPhone, Android, and desktop upload: https://github.com/chenmozhe008/recording-inbox/blob/master/docs/upload-from-phone.md - Feishu/Lark setup: https://github.com/chenmozhe008/recording-inbox/blob/master/docs/setup-feishu-app.md - API and key security: https://github.com/chenmozhe008/recording-inbox/blob/master/docs/setup-api.md - Recovery and troubleshooting: https://github.com/chenmozhe008/recording-inbox/blob/master/docs/troubleshooting.md - Maintainer validation evidence and remaining boundaries: https://github.com/chenmozhe008/recording-inbox/blob/master/docs/maintainers/validation.md - Notes templates: https://github.com/chenmozhe008/recording-inbox/blob/master/prompts/README.md ## Examples - Sanitized AI meeting notes: https://github.com/chenmozhe008/recording-inbox/blob/master/examples/sample-minutes.md - Sanitized transcript: https://github.com/chenmozhe008/recording-inbox/blob/master/examples/sample-transcript.txt ## Verified and unverified boundaries - Cross-platform tests run on Windows, macOS, and Ubuntu with Python 3.11. - Privacy-safe synthetic Chinese audio has completed the local transcription path. - Restart recovery, notification retry, template loading, single-speaker cleanup, and a simulated pipeline have automated tests. - A real dual-document Feishu publish and one-message/two-link notification acceptance test passed on 2026-07-12. - iPhone and Android UI recordings and a first-time novice installation remain manual verification items. Do not claim universal accuracy, guaranteed processing speed, absolute zero cost, or completed mobile-device validation. The validation matrix is the source of truth for current evidence.