Back to search
vedivox
MarketplaceWorks with Claude CodeGitHub
Record and transcribe meetings on your Mac, locally
Safety scan
A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.
Checked today
Badge for your README
Markdown
[](/listings/dimstunt--vedivox)HTML
<a href="/listings/dimstunt--vedivox"><img src="/api/listings/dimstunt--vedivox/badge.svg" alt="findsafeskills safety rating" /></a>Install
Install instructions from the README
As written by the repo's author. findsafeskills has not verified these steps.
brew install --cask dimstunt/tap/vedivox
The cask installs `Vedivox.app` and links the `vedivox` CLI and the `vedivox-mcp` server into Homebrew's `bin`. The Claude Code skills are not part of the cask: they install as a plugin, two commands under [Command line and MCP](#command-line-and-mcp). Your meetings are never removed by `brew uninstall --zap`.
What was scanned
The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.
.claude-plugin/marketplace.json · Manifest · 14 lines
- {
- "name": "vedivox",
- "owner": { "name": "Dmitrii Cherkashin" },
- "metadata": {
- "description": "Claude Code skills for Vedivox, the macOS meeting recorder"
- },
- "plugins": [
- {
- "name": "vedivox",
- "source": "./",
- "description": "Ask about your Vedivox meetings, get help during a call from its live text, and give a meeting a title and a timestamped description"
- }
- ]
- }
README.md · README · 243 lines
- # <picture><source media="(prefers-color-scheme: dark)" srcset="docs/images/vedivox-header-dark.svg"><img alt="Vedivox" src="docs/images/vedivox-header-light.svg" height="64"></picture>
- [](https://github.com/dimstunt/vedivox/actions/workflows/test.yml)
- [](https://github.com/dimstunt/vedivox/stargazers)
- [](LICENSE)
- A macOS meeting recorder that transcribes and tells speakers apart on your Mac.
- If Vedivox helps you, a ⭐ helps others find it.
- 
- 
- Vedivox records your microphone and the system audio of a call (Zoom, Google Meet,
- Microsoft Teams, Yandex Telemost, anything that plays sound), turns both into a transcript, and labels who said
- what. Voices it has met before are recognized by name in later meetings. It
- also types dictation into any app (press a hotkey, speak, press it again), imports existing recordings, and
- exposes everything through a command line and an MCP server, so an assistant
- such as Claude Code can search and summarize your meetings.
- Status: pre-1.0, built for one person's daily use and shared as is. Database
- migrations may still break backward compatibility between versions.
- ## Requirements
- - macOS 14.4 or later on Apple Silicon (the models run on the Neural Engine).
- - Internet on first launch: the speech models are downloaded once (about 1.2 GB).
- After that, recording, transcription and speaker recognition work offline.
- - About 3 GB of free disk for the models and their compiled caches.
- Transcription uses NVIDIA Parakeet TDT 0.6B v3, which covers 25 European languages:
- Bulgarian, Croatian, Czech, Danish, Dutch, English, Estonian, Finnish, French, German,
- Greek, Hungarian, Italian, Latvian, Lithuanian, Maltese, Polish, Portuguese, Romanian,
- Russian, Slovak, Slovenian, Spanish, Swedish and Ukrainian (per the model card); other
- languages are not supported.
- <p>
- <img src="docs/images/sidebar.png" alt="The sidebar: meetings by day, with series and summaries" width="260">
- <img src="docs/images/speakers.png" alt="Speakers: voices Vedivox has learned, recognized by name in later meetings" width="520">
- </p>
- ## Install with Homebrew
- ```bash
- brew install --cask dimstunt/tap/vedivox
- ```
- The cask installs `Vedivox.app` and links the `vedivox` CLI and the `vedivox-mcp` server into Homebrew's
- `bin`. The Claude Code skills are not part of the cask: they install as a plugin, two commands
- under [Command line and MCP](#command-line-and-mcp). Your meetings are never removed by
- `brew uninstall --zap`.
- ## Install from a release zip
- Unpack the archive and run `bash install.sh` in it. The full walkthrough —
- permissions, the CLI and the Claude Code skills — is in
- [docs/INSTALL.md](docs/INSTALL.md).
- ## Build from source
- You need Xcode 26 or later and [XcodeGen](https://github.com/yonaskolb/XcodeGen)
- (`brew install xcodegen`).
- ```bash
- make install TEAM=- # ad-hoc signature, no Apple account needed
- make install TEAM=ABCDE12345 # or sign with your own Apple Developer team
- ```
- `make install` builds a Release `Vedivox.app` into `/Applications` and puts `vedivox`
- and `vedivox-mcp` into `~/.local/bin`. Without `TEAM` the build signs with the
- author's team from `Vedivox/project.yml` when its certificate is in the keychain and
- ad hoc otherwise (the Makefile says so in its first line of output). Dependency
- versions are pinned by the committed `Package.resolved` files.
- With an ad-hoc signature (`TEAM=-`) macOS treats every rebuild as a new app and
- asks for the microphone and system-audio permissions again. A real team ID keeps
- them across rebuilds.
- The same `TEAM=` works for the tests:
- ```bash
- make test TEAM=- # headless logic suite, the gate
- make test-ui TEAM=- # HUD snapshot tests, needs a full app build
- ```
- `Vedivox/Vedivox.xcodeproj` is generated from `Vedivox/project.yml` (`make generate`); do
- not edit it by hand. The benchmark target (`make bench`) and a few recorded
- scenarios rely on private recordings that are not in this repository; those
- tests skip themselves.
- `make dist` packs `Vedivox.app`, the CLI and the docs into a zip for another Mac.
- `make dist-notarized DEVELOPER_ID="Developer ID Application: …"` signs and
- notarizes that zip (see the comment in the `Makefile`); without `DEVELOPER_ID` it takes
- the Developer ID certificate from the keychain.
- ## Command line and MCP
- `vedivox` reads the same database as the app and sends actions to the running app
- over a local Unix socket: `vedivox list`, `vedivox show <id>`, `vedivox transcript <id>`,
- `vedivox search <query>`, `vedivox speakers`, `vedivox import <file|url>`,
- `vedivox transcribe <file|url>`, `vedivox reprocess <id> --yes`, `vedivox record start|stop`,
- `vedivox live` (the text of the meeting being recorded right now),
- `vedivox meeting split|trim|merge|undo` (cut a long recording into calls, join a call recorded in parts),
- and editing commands for speakers, series, marks and segments. The reading commands,
- `transcribe`, `live`, `record`, `meeting split|trim|merge|undo|delete`, `audio compact` and
- `feedback` take `--json`; the rest print text only. `vedivox --help` lists them all.
- Ready-made Claude Code skills live in
- [skills/vedivox](skills/vedivox), [skills/vedivox-live](skills/vedivox-live) (help during a
- call) and [skills/vedivox-enrich](skills/vedivox-enrich) (a title and a timestamped description
- for a meeting). This repository is also a Claude Code plugin marketplace, so the skills install
- and update as a plugin. In Claude Code:
- ```
- /plugin marketplace add dimstunt/vedivox
- /plugin install vedivox@vedivox
- ```
- They are then `/vedivox:vedivox`, `/vedivox:vedivox-live` and `/vedivox:vedivox-enrich`, and
- `/plugin marketplace update vedivox` brings the newer ones after a release. `install.sh` copies
- the same skills into `~/.claude/skills` instead; use one way or the other.
- `vedivox-mcp` is a Model Context Protocol server over stdio. It lists, reads and searches
- meetings (the live transcript of a recording too), imports files and YouTube links,
- reprocesses, splits, trims and merges meetings and undoes those edits, renames and merges
- speakers, reassigns lines, sets a meeting's title, description and series, creates series, and
- adds and removes excluded intervals. It does not record, and has no counterpart of
- `transcribe`, `status`, `queue`, `audio compact`, `feedback`, `mute-debug`, `meeting delete`,
- `speaker detach|delete` or `series set|merge|delete|apply-rules`. Its 22 tools:
- `list_meetings`, `get_meeting`, `get_transcript`, `get_live_transcript`, `search_transcripts`,
- `list_speakers`, `update_speaker`, `merge_speakers`, `update_meeting`, `update_segment`,
- `reprocess_meeting`, `split_meeting`, `trim_meeting`, `merge_meetings`, `undo_meeting_edit`,
- `import_file`, `import_youtube`, `list_series`, `create_series`, `mark_excluded`, `unmark`,
- `list_marks`.
- Point an MCP client at the binary by its full path:
- ```json
- {
- "mcpServers": {
- "vedivox": { "command": "/Users/you/.local/bin/vedivox-mcp" }
- }
- }
- ```
- That is `~/.local/bin/vedivox-mcp` after `install.sh` or `make install`, and
- `/opt/homebrew/bin/vedivox-mcp` (Homebrew's `bin`) after the cask. With Claude Code the
- CLI plus the skill is usually enough, and MCP is optional.
- Who can drive the app: any process running as your macOS user, not just `vedivox`
- and `vedivox-mcp`. The socket does not check which program connects; other users on
- the Mac cannot reach it. Such a process can start a recording and read the
- audio, so Vedivox shows a notification, "Recording started from the command line",
- whenever a recording is started through the socket. See
- [SECURITY.md](SECURITY.md).
- ## Privacy
- Recording, transcription and speaker recognition run on the Mac with Core ML.
- Audio, transcripts, voice profiles and the database stay in
- `~/Library/Application Support/Vedivox/`. There is no account, telemetry, analytics,
- crash reporting or auto-install, and the code calls no cloud or LLM API. The MCP
- server speaks stdio and the app's control channel is a Unix socket readable
- only by you; nothing listens on the network.
- The exceptions, so the claim stays honest:
- - On first launch (and again if the model cache is cleared) the speech models
- are downloaded from Hugging Face over HTTPS. After that Vedivox works offline.
- - `vedivox import <url>` and `vedivox transcribe <url>` run the `yt-dlp` you installed
- yourself, which contacts the video site.
- - Once a day Vedivox asks the GitHub API for its latest release, to tell you when a
- newer version is out. It is one anonymous GET; nothing about you or your
- meetings is sent, and nothing is downloaded or installed. Turn it off in
- Settings → About.
- - `vedivox feedback` writes a zip with logs to your Desktop. Nothing is sent, but
- the logs contain meeting titles, speaker names and file paths — read it before
- you share it.
- - Calendar access uses the local macOS calendar store. Reading the browser's
- tabs (to notice a Google Meet or Yandex Telemost call) happens locally through Apple Events.
- - The app is not sandboxed, so all of the above is a property of the code, not
- something macOS enforces.
- ## Recording other people
- In many countries and US states, recording a conversation requires the consent
- of everyone in it, sometimes announced in advance. Vedivox does not tell anyone
- that it is recording: it shows its own recording indicator on your Mac and
- nothing else, and it records whatever your Mac plays and hears. Knowing the
- rules where you and the other people are, and asking for consent, is your
- responsibility. This is not legal advice.
- ## Documentation
- - [docs/GUIDE.md](docs/GUIDE.md) — the non-obvious parts of the UI: hotkeys,
- modifier clicks, settings files. Also in the app under Help → Vedivox Guide.
- - [CHANGELOG.md](CHANGELOG.md) — what changed in each version.
- - [docs/INSTALL.md](docs/INSTALL.md) — installing a release zip.
- ## Acknowledgements
- Vedivox is glue around other people's work.
- Libraries:
- | Library | Licence | Used in |
- |---|---|---|
- | [FluidAudio](https://github.com/FluidInference/FluidAudio) (includes fastcluster, BSD-2-Clause, and a port of VBx, Apache-2.0) | Apache-2.0 | app |
- | [GRDB.swift](https://github.com/groue/GRDB.swift) | MIT | app, CLI, MCP server |
- | [KeyboardShortcuts](https://github.com/sindresorhus/KeyboardShortcuts) | MIT | app |
- | [swift-argument-parser](https://github.com/apple/swift-argument-parser) | Apache-2.0 with Runtime Library Exception | CLI |
- | [swift-snapshot-testing](https://github.com/pointfreeco/swift-snapshot-testing) and its dependencies | MIT; swift-syntax is Apache-2.0 | tests only |
- Models. None are stored in this repository or in the app bundle; FluidAudio
- downloads them at run time from the Fluid Inference repositories on Hugging
- Face. All of them are Core ML conversions, that is, modified versions of the
- originals.
- | Used for | Downloaded from | Original model | Licence |
- |---|---|---|---|
- | Transcription | [FluidInference/parakeet-tdt-0.6b-v3-coreml](https://huggingface.co/FluidInference/parakeet-tdt-0.6b-v3-coreml) | NVIDIA [parakeet-tdt-0.6b-v3](https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3) | [CC-BY-4.0](https://creativecommons.org/licenses/by/4.0/) |
- | Speaker diarization and voice embeddings | [FluidInference/speaker-diarization-coreml](https://huggingface.co/FluidInference/speaker-diarization-coreml) | [pyannote speaker-diarization-community-1](https://huggingface.co/pyannote/speaker-diarization-community-1), with WeSpeaker embeddings and PLDA from BUT Speech@FIT; the legacy path uses pyannote segmentation-3.0 (MIT, © CNRS) and wespeaker-voxceleb-resnet34-LM | CC-BY-4.0 |
- | Live text while dictating | [FluidInference/Nemotron-3.5-ASR-Streaming-Multilingual-0.6b-CoreML](https://huggingface.co/FluidInference/Nemotron-3.5-ASR-Streaming-Multilingual-0.6b-CoreML) | NVIDIA nemotron-3.5-asr-streaming-0.6b | [OpenMDW-1.1](https://openmdw.ai/license/1-1/) |
- Thanks to NVIDIA, the pyannote and WeSpeaker authors, BUT Speech@FIT and Fluid
- Inference. The licences above are those stated on the model cards in
- November 2026; the cards are the authority if they change.
- The licences and notices of everything built into the app, the CLI and the MCP server, in
- full, are in [THIRD_PARTY_NOTICES.md](THIRD_PARTY_NOTICES.md), which also ships in the
- release zip.
- URL import relies on [yt-dlp](https://github.com/yt-dlp/yt-dlp) and
- [ffmpeg](https://ffmpeg.org), which Vedivox does not ship: install them yourself
- (`brew install yt-dlp ffmpeg`) if you want that feature.
- ## Licence
- Vedivox is free software under the [GNU General Public License, version 3 or any
- later version](LICENSE) (GPL-3.0-or-later). © 2026 Dmitrii Cherkashin
- Issues, ideas and pull requests are welcome; a pull request is merged under the
- contributor licence agreement in [CONTRIBUTING.md](CONTRIBUTING.md).
Ratings and comments
Sign in to rate this skill.
Sign inFree to use. If it’s helped, pass it on.