Vecta 0.1.0 — local macOS developer preview (build 65)

Vecta 0.1.0 — developer preview (build 65)

A native Apple Silicon macOS app for keeping PLAUD recordings and processing them locally.

• Local transcription, saved speaker profiles, text summaries and visual summaries. No hosted inference or cloud fallback.
• Independent direct BLE and local Wi-Fi Fast Transfer. No PLAUD SDK, relay website or intermediary recording-transfer service.
• Automatic synchronization and local processing while Vecta is open. It does not keep running after you quit the app.
• PDF and TXT exports of saved results, with native visual-report layout.
• English and Russian interface.
• 8 GB unified memory minimum for Qwen 4B; 16 GB for Qwen 9B. Recording length, free memory and disk space still matter; this is not an 8 GB workload benchmark.

Setup and installation

Apple Silicon, macOS 15 or later. Model weights are not bundled; installation is an explicit action in the app. Internet is needed for initial authorized PLAUD Partner provisioning and model downloads, and may be needed for a later requested authorization refresh or another model installation. Normal configured transfer and processing can work without Internet.

Modern recorder onboarding requires your own authorized PLAUD Partner credentials; no shared keys or Partner access are included. Local audio import does not need device credentials. See the README and installation guide.

Ad-hoc signed, not Developer ID signed, and not notarized. A clean-user downloaded/quarantined-DMG first launch has not been tested. Verify the origin and SHA-256 before installation, back up an existing library, and use Apple's documented Privacy & Security → Open Anyway route only if you trust the app. Do not disable Gatekeeper or bypass a malware warning.

Verification and matching source

Built from commit 6bb777beaf9535fb769221e479328a421c6939d2.

Independent full CI also passed all three gates: source/security, model-free macOS regressions and the complete ARM64 Speech/MLX app package.

Clean hosted release preparation passed source/security, packaging, native executable regressions, 55 Speech configuration checks, 153 transport tests, 18 MLX tests, actual LAME relinking, production bundle verification and mounted-DMG/copy verification. No model weights were downloaded or real recordings processed by these checks. The downloaded release source archive was independently compared byte-for-byte with all 234 reviewed files, and the DMG was independently verified and mounted read-only without launching or installing it.

DMG SHA-256:

920befcfc88f180e42458b7254428227c5d56e86d69a670cf5207b5da88c7d8b

Download the DMG together with its matching source archive, component-license archive and LAME source/object/relink kit. SHA256SUMS covers the build's five primary files. artifact-verification.json records the packaging checks at build time; its published: false field describes the preparation stage, not the later release publication.

Screenshots and the sample PDF in the repository use fictional prepared meeting data, not owner recordings or model-quality benchmarks. Physical acceptance is limited to the documented owner NotePin S/Mac setup; every firmware/device, minimum-OS installation and long-recording quality are not claimed.

Report security concerns through GitHub Private Vulnerability Reporting. Do not submit private recordings, credentials or exploit details publicly.