# Vox.md > Vox.md is an open-source, local-first quick-capture app for iPhone, iPad, Mac, and Apple Watch. It captures text, links, photos, files, scans, sketches, and voice, then writes standard Markdown and attachments into a user-selected Obsidian vault or Files folder. Voice transcription and supported AI processing run on the Apple device; capture content is not sent to Vox.md servers. Canonical product name: Vox.md Canonical website: https://vox.isolated.tech/ App Store: https://apps.apple.com/us/app/voxboard/id6758967337 Source: https://github.com/CodyBontecou/Voxboard URL scheme: voxboard:// Documentation updated: 2026-08-22 ## Documentation - [Complete Feature Guide](https://vox.isolated.tech/docs/): Every shipped user-facing feature, organized by category. - [Getting Started](https://vox.isolated.tech/docs/#getting-started): Platforms, minimum versions, setup, permissions, and first capture. - [Quick Capture](https://vox.isolated.tech/docs/#quick-capture): Durable composer, capture types, Capture Bar, inline voice, and input limits. - [Capture Presets and Routes](https://vox.isolated.tech/docs/#presets): Processing modes, metadata, Markdown destinations, placement, templates, overrides, retry protection, and audio rules. - [Current Location, {location}, and Metadata](https://vox.isolated.tech/docs/location/): Per-preset acquisition, token-only links, optional metadata output, precision, capture-source timing, migration, privacy, and troubleshooting. - [Models and Transcription](https://vox.isolated.tech/docs/#models): Automatic Apple Speech, Whisper, Parakeet, languages, downloads, and pause detection. - [iOS Keyboard](https://vox.isolated.tech/docs/#keyboard): Full Access setup, dictation, live insertion, fallback behavior, haptics, and auto-stop. - [System Integrations](https://vox.isolated.tech/docs/#system-integrations): Share Extension, widgets, Lock Screen, Control Center, Live Activities, Shortcuts, and deep links. - [Apple Watch](https://vox.isolated.tech/docs/#apple-watch): Local recording, preset selection, iPhone queue, recording-only output, and complication. - [Mac Companion](https://vox.isolated.tech/docs/#mac): Capture workspace, meeting capture with separate system and microphone stems, audio import, menu bar, visibility, global keybinds, history, and models. - [History, Settings, Pricing, and Privacy](https://vox.isolated.tech/docs/#history-settings): Local history, recovery, private activity stats, settings map, free allowances, lifetime unlock, and data handling. - [Troubleshooting](https://vox.isolated.tech/docs/#troubleshooting): Destination, keyboard, model, transcription, heading, and Watch queue issues. ## Product Facts - Minimum versions: iOS/iPadOS 17.6, macOS 14, and watchOS 10. - Control Center controls require iOS 18. - Automatic Apple Speech requires a supported iOS 26 device and language. - Apple Intelligence processing requires compatible iOS/macOS 26 hardware. - No account or Obsidian plugin is required. - Destinations are user-selected Obsidian vaults or other writable Files folders. - Note targets are new notes, rolling notes, and existing Markdown notes. - Rolling periods are daily, weekly, monthly, quarterly, and yearly. - Placement options are top, bottom, or beneath a level 1–6 heading; missing headings can error or be created. - Capture Preset processing modes are Keep Original, Clean Prose, Todo Checklist, Meeting Notes, and Custom Instruction. - When Apple Intelligence is unavailable, Todo and Meeting Notes have deterministic formatters; Clean Prose and Custom Instruction preserve the original text. - Typed and mixed-media processing is opt-in per preset. Voice transcripts use the selected preset mode. - Static metadata can be note-level frontmatter or per-entry inline fields. - Per-preset location metadata is opt-in. Each Capture Preset has two independent location choices. Use Current Location is the explicit acquisition opt-in and powers `{location}`. Write Location Metadata is a separate output opt-in for frontmatter or inline fields. Operating-system permission is only a gate and never opts a preset in. - Both controls are off for new and previously disabled presets. Presets that already had location enabled before the controls were split keep metadata output enabled after migration, preserving their prior behavior. - A preset with Use Current Location enabled requests one origin-time location at Send for iOS/iPadOS, Share Sheet, deep-link/composer, Mac, and system actions that open the composer; during invocation for direct-enqueue text, link, and file intents; or at recording stop for shortcut/widget/control recordings, immediate voice, keyboard-driven preset recording, and Watch captures routed through normal Markdown delivery. Watch Recording Only is a raw audio Files export with no Markdown surface and skips location. There is no continuous or background location tracking. - Entry prefixes, suffixes, reusable Entry Templates, and vault Markdown templates can use `{location}` for a Google Maps Markdown link even when Write Location Metadata is off. When Current Location is off or unavailable, the token renders empty. The composer detects the token in resolved inline entry formatting and offers a one-tap fix that enables acquisition while leaving metadata output off; live vault template files are read only at delivery and require manual preset setup. - Exact is the default location precision. City rounds both coordinates to two decimal places and omits point-of-interest labels; every derived field, `{location}` link, and provider link uses the same privacy-adjusted coordinates. - Location metadata supports renameable structured fields or validated advanced YAML. Structured fields can be document frontmatter or per-entry inline fields. Document frontmatter retains multiple locations in an idempotent collection keyed by Capture ID; advanced YAML cannot replace unrelated frontmatter. - Apple’s system reverse geocoder is used only when metadata output is enabled and selected fields need labels. Token-only location skips reverse geocoding. The geocoder is free and requires no API key, may use the network, and fails softly to coordinate-only metadata. Provider links disclose their exact or rounded coordinates to Apple, Google, or OpenStreetMap only when the user opens them. - If location is unavailable, interactive surfaces can retry, cancel, send once without it, or persist Always Send Without Location per preset; preset settings expose a reset. An unattended Ask result is durable and never reacquires a later location. - Reusable Entry Templates, vault template files, prefix/suffix text, path tokens, attachment subfolders, and retry markers are supported. ## Capture Inputs and Editor Quick Capture accepts: - Typed Markdown and pasted text. - HTTP/HTTPS links. - Photos, selected screenshots, and camera images. - Multiple arbitrary files. - Multipage document scans with PDF generation, page images, and best-effort on-device OCR. - Up to 10 manually captured journal pages or selected journal photos with on-device OCR. - PencilKit sketches with editable drawing data and PNG preview. - An independent explicit one-shot Current Location Capture Bar action that inserts a Google Maps Markdown link; it does not change either per-preset location control. - Voice recordings and imported audio/video, transcribed locally. The configurable Capture Bar includes Add Media, Add Files, Scan Document, Extract Text, Undo, Format Markdown, Markdown Link, Set Due Date, Checklist, Bullet List, Paste, Internal Link, Sketch, Current Location, Insert Timestamp, Insert Date, and Change Text Case. Actions can be reordered or hidden. Shared request limits are 100,000 text characters, 10 shared items, and 250 MB aggregate staged assets. Overflow is rejected instead of silently truncated. Inline voice offers Add to Draft and Send Immediately. Send Immediately is selected by default in the iOS and Mac capture workspaces. The iOS `voxboard://capture?action=voice` entry point switches its recording to Add to Draft. ## Transcription Backends Automatic is the iOS default. It prefers Apple's system-managed, on-device speech recognizer on supported iOS 26 devices and falls back only to a local model the user explicitly downloaded. Optional local models: - Whisper Tiny: 75 MB. - Whisper Base: 142 MB. - Whisper Small: 466 MB. - Whisper Medium: 1.5 GB. - Whisper Large v3 Turbo: 1.6 GB. - Parakeet v2: approximately 800 MB, optimized for English. - Parakeet v3: approximately 800 MB, supports 25 languages. Whisper uses whisper.cpp. Parakeet uses FluidAudio and Core ML. Model downloads are user-initiated and removable. Transcription runs locally after preparation. Explicit Whisper and Parakeet use batch transcription; Automatic can expose finalized and tentative live Apple Speech text. ## Keyboard Safety Note The current Vox.md iOS keyboard requires Apple's Allow Full Access setting so the keyboard can use shared App Group state and coordinate recording with the main app. Vox.md does not upload keystrokes, dictated text, transcripts, or microphone audio to Vox.md servers. Do not claim that the current keyboard works without Full Access. ## Apple System Entry Points App Intents include Record with Vox.md, Open Quick Capture, Record a Capture, Capture a Screenshot, Capture Text to Markdown, Capture Link to Markdown, and Capture File to Markdown. Deep-link examples: - voxboard://capture - voxboard://capture?action=photos - voxboard://capture?action=screenshots - voxboard://capture?action=camera - voxboard://capture?action=files - voxboard://capture?action=scan - voxboard://capture?action=sketch - voxboard://capture?action=link - voxboard://capture?action=voice - voxboard://capture?text=URL_ENCODED_TEXT - voxboard://capture?url=URL_ENCODED_HTTPS_URL - voxboard://capture?preset=PRESET_ID - voxboard://listen Unknown or duplicate parameters, malformed identifiers, non-HTTP URLs, and oversized text are rejected. ## Privacy and Local Data Capture drafts, attachments, audio, transcripts, templates, routes, file paths, OCR results, typed text, and keyboard input are handled locally and are not sent to Vox.md servers. Files are written only to the destination selected by the user. If that folder uses iCloud Drive or another provider, that provider can sync it under its own terms. Production onboarding analytics are disabled by default. Vox.md has no third-party analytics SDK, ad network, or cross-app tracker. Location is not collected by the developer or analytics. User-initiated model downloads contact external model-distribution infrastructure, currently Hugging Face directly or through FluidAudio, which can receive ordinary connection metadata but no capture content. Pending and failed requests may retain the privacy-adjusted location snapshot or unavailable outcome needed for recovery. After successful delivery, recoverable Capture requests and completed Watch queue items are reduced to scrubbed tombstones. Completed Capture history stores coarse delivery metadata, not note content, URLs, coordinates, location labels or templates, bookmarks, absolute paths, or attachment filenames. - [Privacy Policy](https://vox.isolated.tech/privacy.html): Complete data handling and retention details. - [Terms of Service](https://vox.isolated.tech/terms.html): Product terms and transcription disclaimers. ## Pricing The free tier includes 15 minutes of transcription and 10 successful Capture deliveries. Failed writes and idempotent retries do not consume a successful delivery. Unlimited use is a one-time lifetime purchase with no subscription or renewal. Localized App Store pricing is authoritative; the app uses $9.99 as a fallback display price. ## Additional Resources - [Homepage](https://vox.isolated.tech/): Product overview and screenshots. - [Blog](https://vox.isolated.tech/blog/): Capture and private dictation guides. - [GitHub README](https://github.com/CodyBontecou/Voxboard#readme): Technical overview and repository map. - [Setup](https://github.com/CodyBontecou/Voxboard/blob/main/SETUP.md): Local development setup. - [Contributing](https://github.com/CodyBontecou/Voxboard/blob/main/CONTRIBUTING.md): Contribution workflow. - [Unified Capture Architecture](https://github.com/CodyBontecou/Voxboard/blob/main/docs/unified-quick-capture-destinations.md): Design history; distinguish planned sections from shipped behavior. - [App Store](https://apps.apple.com/us/app/voxboard/id6758967337): Current distribution and localized purchase details. ## Scope Notes for Agents - Use “Vox.md” in user-facing text. “Voxboard” remains in source paths, bundle identifiers, the GitHub repository name, and the voxboard:// URL scheme. - The complete feature guide documents reachable, shipped product behavior. Do not treat dormant legacy screens or internal Smart Folder flags as shipped features. - Speaker diarization is an opt-in Capture Preset feature for supported voice recordings and imported audio on iPhone, Apple Watch transcript delivery, and the native Mac app. It adds anonymous best-effort labels locally after first-use model preparation; failures preserve the plain transcript and show a reason in History. - Capture Preset is the current name for the reusable workflow formerly represented by legacy internal “Vox” types. - Prefer the code-backed documentation above over stale marketing claims or planning documents.