Skip to main content

Reference

Changelog and roadmap

A high-level record of what shipped and what's coming next. Format follows Keep a Changelog loosely — focused on what users notice, not internal refactors.

June-July 2026 updates

  • Cloud AI tiering changed. Free summary providers are Built-in AI (on-device on macOS, Windows, and Linux) and Ollama on every platform. Mistral, OpenAI, and Anthropic cloud providers require Pro, even with your own API key.
  • Managed keyless cloud remains roadmap. The current Pro cloud path is BYOK. Managed Mistral or Anthropic access is planned for v2.1 and is not live.
  • Study Sage mode gates shipped. Chat stays Free. Quiz, Explain, and Challenge require Pro.
  • Gemini is deprecated. It is legacy-only for existing saved keys and is not offered for new setup.
  • Speaker diarization shipped with limits. Supported builds expose Pro-gated diarization with downloadable models and You/Remote channel labels where available.
  • Silero neural VAD shipped as optional. RMS VAD remains the default. Silero is Free when compiled in, default off, and requires a model download plus the setting toggle.
  • React 19 and Next 15.5. The desktop app frontend now runs on React 19 and Next 15.5.
  • macOS builds are Apple Silicon only. Intel Mac users are not covered by the shipped macOS binary.

v2.0 baseline (April 2026)

The v2 line is a ground-up rewrite of NeuroBridgeEDU on Tauri 2. The headline values: privacy-first, neurodivergent-first, fully offline-capable, with an opt-in Pro path to cloud AI providers.

Recording & audio

  • Microphone capture via cpal.
  • System audio capture via WASAPI (Windows), virtual audio device (macOS), and PulseAudio / PipeWire loopback (Linux).
  • RMS-based ducking and soft-limiting / anti-clipping in the recording path.
  • RMS-energy VAD on the transcription path (10–60% workload reduction).
  • Audio file import via Symphonia for MP3, M4A / AAC, FLAC, OGG, WAV.

Transcription

  • Whisper.cpp via whisper-rs with GPU acceleration (Metal / CUDA / Vulkan, selected at build time).
  • Single-pass inference with internal ~30-second windowing; segment results stream back to the UI as they're produced.
  • Whisper model catalog: tiny, base, small, medium, large-v3, large-v3-turbo, plus quantised q5_0 variants.

AI summaries

  • Current setup providers behind a common trait: Built-in AI (on-device on macOS, Windows, and Linux, via llama-cpp-2), Ollama, and Pro cloud BYOK providers Mistral, OpenAI, and Anthropic. Gemini is legacy-only for existing saved keys. Custom OpenAI-compatible endpoints are deferred.
  • Built-in templates: Meeting Summary, Lecture Notes, Study Session, Brief, Detailed, Research Study, Technical, Action Items.
  • Custom user templates with arbitrary prompts in Settings → Customisation → Create Custom Template.
  • Automatic offline fallback to Built-in AI (on-device on macOS, Windows, and Linux) when cloud providers are unreachable.

Study Sage

  • Chat over your meeting transcripts and summaries.
  • Chat is Free. Quiz, Explain, and Challenge require Pro.
  • Cmd+Shift+K / Ctrl+Shift+K shortcut.
  • Context retrieval via SQLite FTS5 + BM25.
  • Conversations persist locally so they survive an app restart.

Pro features (require Pro account)

  • Recall Engine — flashcards generated from meeting summaries.
  • Exam Radar — AI analysis over stored summaries with Student / Lecturer / General modes.
  • PDF export — typeset PDF of meeting summaries via Typst.
  • Action items extraction — Tasks / Decisions / Open Questions extracted from a summary on demand.
  • Analytics — aggregate stats over your meetings.
  • Summary preset packs — extra prompt presets persisted on Pro accounts only.

PDF export

  • Markdown summary → Typst markup → PDF, all in pure Rust (no external deps).
  • Embedded NeuroBridgeEDU logo, page-numbered footer, brand line.

Search

  • SQLite FTS5 + BM25 ranking over titles, transcripts, and summary text.
  • FTS5 powers Study Sage's content retrieval.
  • The user-facing meeting search box does title-only substring matching (FTS5 for the user search is on the roadmap).

What's coming next

Near term — v2.1

Theme: better defaults, fewer setup steps.

  • Managed keyless cloud for Pro, so users can choose a hosted Mistral or Anthropic path without managing their own API key.
  • macOS native system audio via ScreenCaptureKit, removes the BlackHole / Loopback requirement.
  • Saved searches in the search panel.
  • Card import / export for Recall Engine — bring in or share Anki decks.
  • Bulk PDF export — multiple meetings as a single document.
  • Transcript regeneration that also re-runs the summary in one click.
  • Locale fan-out: translate the in-app Help topics added in v2.0 into the 32 supported UI languages (English-only for now).

Mid term — v2.2 / v2.3

Theme: smarter defaults, smarter recall.

  • Real-time translation — transcribe in source language, translate to target inline.
  • Cross-meeting semantic search — vector-based retrieval beyond lexical FTS5.
  • Topic clustering in Analytics — group related concepts automatically.
  • Comparison views in Analytics — week-over-week, semester-over-semester.
  • Multi-device sync for Recall Engine (opt-in, end-to-end encrypted).
  • Custom Typst templates for PDF export, configurable from the UI.
  • Recall Engine spaced repetition — persistent flashcards with an SM-2-style scheduler so cards survive sessions and return at smart intervals.

Longer term — v3.x

Theme: collaborative learning, beyond the desktop.

  • Mobile companion app for review-only (flashcards, summaries) on iOS / Android. Recording stays desktop.
  • Optional shared spaces — opt-in encrypted sharing of summaries (not recordings) with study partners.
  • Plugin system — third-party summary templates and chat tools.
  • Custom local LLM management — easier model swaps, fine-tunes, and quantisation choices.

Things we've decided not to do

Sometimes the most useful thing is to be clear about what's out of scope.

  • A cloud-hosted version — the core value is local-first. We won't host your data.
  • Mandatory accounts — there's no sign-in for desktop and there won't be one.
  • Always-on telemetry — error reporting will always be opt-in.
  • Video recording — audio-only is a deliberate scope. Use OBS or your platform's built-in recorder if you need video.
  • Re-implementing Anki — Recall Engine intentionally stays simple. If you outgrow it, export to Anki via the planned import / export feature in v2.1.

For the latest commit-level history, see the GitHub commit log.

Edit this page on GitHub