Your PLAUD recordings, directly transferred and processed on your Mac.
Vecta is a native Apple Silicon macOS app that connects to compatible PLAUD recorders over Bluetooth and local Wi-Fi Fast Transfer — without the PLAUD SDK, a relay website, or a recording-transfer middleman. Keep recordings in a searchable local library, transcribe and summarize them on your Mac, remember speakers across meetings, and export visual or PDF reports when you need to share the result.
Recording content is not sent to cloud inference, and Vecta has no cloud-inference fallback. The interface supports English and Russian.
Download Vecta 0.1.0 build 80 · Installation · Build from source
Screenshots show the real native interface with fictional meeting data. The demo transcript and summaries are prepared examples, not model-quality benchmarks.
Vecta implements its own BLE and local Wi-Fi Fast Transfer, without the PLAUD SDK, a relay website, or an intermediary service for recording transfer. Audio is received directly from the selected recorder and kept in your local library.
Speech recognition, speaker separation, text and visual summaries, and the assistant run on your Mac. No recording content is sent to cloud inference, and there is no cloud fallback. You do not need Docker, a browser dashboard or a separate model server.
Internet access is needed for initial device authorization and model downloads. Once your recorder is configured and the models are installed, normal recording transfer and processing work without Internet access. Installing another model or deliberately refreshing authorization may require it again.
Device setup: first-time connection to a modern PLAUD recorder requires your own authorized PLAUD Partner credentials. Vecta does not include shared keys or provide Partner access. You can import local audio without device credentials. See setup and network boundaries.
Save named voice profiles and use them in later recordings. Confirm a suggested speaker, correct a name, or merge speaker groups while preserving saved results. Voice samples and profiles remain local.
Read the main points as a visual report: decisions, responsibilities, next steps, quotes and relationships. Text summaries and the visual report stay with the recording, alongside the original transcript and processing history.
Export a transcript, text summary, visual summary or complete report as PDF or TXT. PDF export preserves the visual layout and uses saved results; exporting does not start another AI job.
Open the fictional meeting's sample PDF.
Choose a sync interval and enable automatic summaries. While Vecta is open and the recorder is available, new recordings can be downloaded, transcribed and summarized locally without starting each job by hand. You choose the summary formats and can pause or cancel processing.
Automatic sync runs while the app is open; Vecta is not a background service that continues after you quit it.
Opening Conferences does not request Screen Recording or microphone access. Recording permissions are checked when you choose Start; unattended conference recording waits for existing permissions instead of prompting in the background. An already authorized microphone is not requested again.
Build 79 uses a stable Apple Development signing identity rather than a different ad-hoc code identity on each rebuild. Switching identities may require a one-time new macOS grant. This does not override revoked permissions, OS reminders or Gatekeeper, and is not Developer ID signing or notarization.
Launch at login is on by default. On the first normal launch, Vecta asks macOS to register it as a login item; macOS may require approval. The setting shows the current status and how to continue.
Keep running in background is also on by default. Closing the main window keeps Vecta and its work open; reopen it from the Dock, Settings or the status menu. Choose Quit Vecta or press Command-Q to terminate it. Turn the setting off if closing the window should quit the app.
Fast transcription is on by default. While Vecta remains open—including when its window is closed in background mode—it may retain roughly 700 MB of additional RAM so later work can start sooner. The first preparation can still take time. Turn it off to use the earlier per-recording transcription mode.
The Formats sidebar lets you create and edit custom text-summary instructions, then select them for manual summaries, automatic processing and Conferences. Deleting a custom format is permanent for that definition; saved results and queued recipe snapshots that used it are retained.
Record system audio, a Mac microphone, both together, or a selected running application for a conference. Recording is opt-in: choose Start, or explicitly enable automatic conference recording in Settings. macOS requests the needed microphone or system-audio permission for the selected source. The optional menu-bar control provides source selection and Start/Stop without recording automatically.
Keep recordings, transcripts and different summary versions together. Search the library, listen to the original audio, ask the assistant about a recording, or ask questions across saved transcripts.
- Apple Silicon Mac, macOS 15 or later. Intel and iOS/iPadOS are not supported.
- 8 GB unified memory for Qwen 4B; 16 GB for Qwen 9B. Available memory and recording length still matter. The minimum is not a guarantee for every workload.
- Several GB of free disk space for models, recordings and processing files.
- Models are installed separately through an explicit user action.
Speech recognition uses either WhisperKit large-v3 Turbo (default; English and other languages) or Sber's GigaAM v3 on MLX (Russian speech, punctuated), chosen in Settings → Local AI. Both engines share SpeakerKit speaker separation and every downstream feature. Summaries and assistant features use MLX Qwen3.5 4B/9B. The interface is available in English and Russian; new processing jobs retain their selected language policy.
Download the DMG from Releases, open it, and drag Vecta to Applications. Back up an existing library before replacement or migration.
Build 79 is a developer preview signed with Apple Development, not Developer ID, and is not notarized. Gatekeeper may block it. Verify the release origin and checksums; if you choose to trust it, use Apple's documented Privacy & Security → Open Anyway route. Do not disable Gatekeeper or bypass a malware warning. See installation instructions.
A clean-user, downloaded-DMG first-launch test is not claimed for this preview. Verification status and limits distinguish local, GitHub CI, physical-device and clean-user checks.
- PLAUD NotePin S: direct BLE and Fast Transfer accepted on an existing owner recorder/Mac setup, including MP3 import and previous-network restoration.
- PLAUD Note, NotePin and Note Pro: adapters are present; compatibility with every device and firmware is not claimed.
- Local audio: import recordings without a PLAUD account.
Vecta is independent software, not an official PLAUD product.
Initial Partner provisioning sends authorization and device metadata to PLAUD Partner endpoints. It does not upload recording audio, transcripts, summaries, voice embeddings or documents. Credentials and cached device identities are kept in the login Keychain; valid cached reconnection is local.
Model installation downloads pinned revisions from Hugging Face and verifies file sizes and SHA-256. Inference workers are launched with a deny-network policy. Fast Transfer uses the recorder's local Wi-Fi network and restores the previous network afterward. It does not route audio through an Internet service.
The main app is not advertised as App-Sandbox-entitled. It uses macOS permissions, filesystem access and bookmarks; local inference workers have a separate network-denying sandbox policy. Read the security policy before reporting a vulnerability. Never attach private recordings or credentials to a public issue.
Install full Xcode 26.4+ with Swift 6.3+/Metal tools, Python 3.9+, Git, make and curl. From the source root, using a new output directory:
python3 -I Tools/release.py build dist/Vecta-0.1.0 --version 0.1.0 --build 79The command builds and checks the native app, workers, codecs, relink kit and DMG. It does not install an app, download model weights, operate a recorder or publish anything. No CMake, llama.cpp, PLAUD SDK or hosted backend is required.
See BUILD, CONTRIBUTING, CHANGELOG
and source provenance. Main branch: main; development:
develop. Preserve user data, historical schemas and third-party notices.
Vecta is licensed under Apache-2.0. It retains adaptations from the Apache-2.0 PinPoint baseline, with NOTICE, provenance and import/rename ledgers. It makes no clean-room claim.
Dependencies and model weights have separate licenses. Each distributed DMG is accompanied by matching source, component notices and a working LAME source/object/ relink kit. See dependency obligations.