Features

Features below are grouped by what you reach for them for, not by where they live in the menu. Pro features are marked. Everything not marked Pro is in the free version.

Capture

Recording

One-shortcut start: Cmd+Shift+R (macOS) or Ctrl+Shift+R (Windows / Linux). Captures microphone and (optionally) system audio simultaneously. Pause / resume with Space; stop with the same shortcut or Escape.

The recording surface shows a live audio meter (so you can confirm you're being heard), a running timer, and the live transcript scrolling in as you speak.

Transcription

Whisper.cpp running locally with GPU acceleration where available (Metal on macOS, CUDA on NVIDIA, Vulkan on AMD/Intel). 50+ languages with auto-detect. Multiple model sizes from tiny (~39 MB) up to large-v3 (~2.9 GB) — pick the trade-off between speed and accuracy that suits your hardware.

Voice Activity Detection (Silero VAD) strips silence before transcription, cutting processing time by 10–60% on real-world recordings. Long recordings stream segment by segment so you can read what's already done while the rest catches up.

Import

Already have a recording? Import MP3, M4A / AAC, FLAC, OGG, or WAV files. Symphonia handles the decoding; the file joins the same transcription pipeline as a live recording, so the output and the downstream features are identical.

Summarise

Local summarisation (Built-in AI)

Mistral and EuroLLM models running on-device via llama.cpp. The default for new installs — no internet required, no data leaves your machine. Choose your tier based on hardware:

• Ministral 3B Instruct (~2 GB) — the recommended starting point. Runs comfortably on 8 GB RAM.
• Ministral 8B Instruct (~4.6 GB) — sharper reasoning for dense, multi-topic recordings. 16 GB RAM recommended.
• EuroLLM 9B (~5.3 GB) — best for European language mixing and code-switching. 16 GB RAM recommended.
• Mistral Nemo 12B (~7.1 GB) — highest local quality for long, complex lectures. 24 GB RAM recommended.

Cloud summarisation (BYOK or Pro)

Bring-your-own-key for OpenAI, Anthropic Claude, Google Gemini, Mistral Cloud, or a self-hosted Ollama server. Pro adds managed access to Mistral Cloud and Anthropic without the BYOK setup. Cloud is opt-in per session — your default stays local until you choose otherwise.

Reading-style preferences

Three densities, configurable globally and per session:

• Just the highlights — skim-friendly bullets, ~30-second read.
• Good balance — structured notes with key terms, ~2-minute read.
• Full picture — long-form recap with context, ~5-minute read.

Behind a "+2 More preferences" expander you can also tune structure (bullets / mixed / paragraphs) and visual emphasis. Defaults are sensible — most users don't need to open it.

Recall (study-time features)

Study Sage chat (Pro)

Conversational AI grounded in the transcript and summary of your recordings. Open with Cmd+Shift+K / Ctrl+Shift+K. Four modes:

• Quiz — generate questions to test recall.
• Explain — re-explain a concept in plainer language.
• Challenge — push back on what you think you know.
• Chat — open conversation about the material.

Recall Engine (Pro)

Auto-generated spaced-repetition flashcards from your summaries. Cards are scheduled by the SM-2 algorithm; review surfaces in a focused study view that respects reduced motion.

Exam Radar (Pro)

Cross-meeting concept analysis. Looks across an entire semester of recordings and surfaces the topics that recur most often, the concepts most likely to appear on an exam, and the gaps in your existing coverage. Useful in the last two weeks of a module.

Action items and decisions (Pro)

On-demand extraction of Tasks, Decisions, and Open Questions from a meeting. Useful for meeting recordings; less useful for solo lectures.

Search and analytics

Full-text search

SQLite FTS5-backed search across every transcript and every summary. Searches in real time as you type. Filter by date, by language, by speaker (Pro), or by tag. Click a result to jump to the exact sentence in the source recording.

Lecture analytics dashboard (Pro)

Topic coverage over time, duration trends, keyword frequency, speaker identification, and an at-a-glance health view of your recording practice across a term.

Export

PDF export (Pro)

Render any meeting as a formatted PDF: summary, transcript, action items, timestamps, and any diagrams the model produced. Renders via the Markdown → Typst → PDF pipeline (Typst is much faster and cleaner than a headless-Chrome alternative).

Plain-text and Markdown

Copy any summary or transcript as plain text or Markdown — free, no subscription. The format you pick matches what most note-taking apps expect.

Accessibility features (always on)

• Full keyboard navigation with a Command Palette (Cmd+K / Ctrl+K).
• Screen reader semantics — native landmarks, no DIY ARIA widgets.
• Respects prefers-reduced-motion system-wide.
• Low-stim mode desaturates colour and disables ambient motion entirely; focus mode hides chrome so the recording surface is the only thing on-screen.
• 30+ UI languages.
• Lexend as the body face — designed for reading proficiency, particularly helpful for dyslexic users.

For deeper detail on any of these, see the corresponding markdown file in the v2 app's documentation/03-features/ folder.