01
Pick a screen
Choose the display that contains the work you want help with. Screenshots are attached only when your request uses screen context.
Private context assistantmacOS · Windows · Linux
Your desktop has the context.
You choose what AI sees.
Volyx Lens brings your screen, voice, and meeting audio into one compact assistant. Select the inputs you need, choose your response and transcription providers, and keep control of the route.
Windows and Linux builds are live in v0.7.4, and unsigned macOS DMGs are attached to the release; the README gives the first-launch steps. The macOS build will be signed with an Apple Developer ID and notarized once release signing credentials are configured. Bring your own provider key.
Context aperture
Ready
Selected context
Screen + microphone
Your provider
3context inputs
11response providers
0Volyx Lens relay servers
1compact desktop rail
Explicit captureProvider choiceLocal credentialsVisible state
Context, on your terms
A lens should focus.
Not vacuum up everything.
Volyx Lens separates context into explicit inputs. Turn on only what the task calls for, then see the active combination before you send a prompt.
02
Add your voice
Speak naturally instead of switching windows to type. The microphone is a separate input with its own visible state.
03
Include system audio when needed
Bring in the other side of a call for meeting or interview context. System audio stays optional and independent.
A route you can inspect
Your context takes the short path.
Volyx Lens sends selected context directly from the app to the AI provider you configure. There is no Volyx Lens-operated intermediary server in that request path.
- 01Provider credentials use Electron secure storage backed by the operating system (macOS Keychain, Windows DPAPI, Linux libsecret) when available. If secure storage is unavailable, keys are kept in memory for the session only and are never written to disk; Settings says so.
- 02The app has no analytics or advertising SDK.
- 03You can switch providers without changing the workflow.
- 04The active transcript and Task Context stay memory-only, within fixed size limits, for the current session, and are cleared when the session ends. Saved meeting history and crash recovery are on by default for new installs; transcript text is written to this device only, and you can turn both off at any time.
Your chosen AI provider still receives the context you send and applies its own retention and privacy terms. Review those terms before sharing sensitive material.
Your computer
Selected context
ScreenMicAudio
Volyx Lens
Local app
Your choice
AI provider
No Volyx Lens cloud relay
Built for the edge of your workspace
Present when you need it.
Compact when you do not.
The floating rail can move around your desktop. Expand it into chat, choose the context for the moment, then collapse it back out of the way.
01Choose a provider
02Fallback, never a blended answer
One assistant, different context
Keep the task in view.
01
Work through screen and coding context
Ask about an error, code, a dense document, a dashboard, or an unfamiliar interface without restating every visible detail.
Screen → Prompt
02
Build Task Context across screens
Save multiple selected screens, such as files, visible problems, and test output, in local memory. A later explicit request uses a bounded visual window and discloses omitted screens.
Multiple screens → Prompt
03
Keep up with a live conversation
Use separate You/Them transcript channels during a permitted meeting, then explicitly request reply suggestions, follow-up questions, or meeting recaps.
You + Them → Prompt
04
Detect and save the meetings you own
Opt-in meeting detection looks for a sustained two-sided conversation and tags the session locally. Meeting history is on by default for new installs: saved meetings get a meeting badge, stay searchable, and export as records, transcripts, or generated meeting notes.
Meeting → History
Bring your own model access
Eleven response providers.
No forced route.
Configure the response and transcription providers that fit your work. Model availability and charges come from those providers, not Volyx Lens.
Realtime transcription runs on OpenAI, Azure Foundry, Deepgram, or Azure AI Speech. Batch transcription falls back to OpenAI, Google Gemini, or Groq Whisper. Local Whisper works only when a whisper-server binary is configured, and the published build does not bundle one. Pick a transcription language or let it auto-detect.
Response providers
- OpenAI
- Anthropic
- Google Gemini
- Azure Foundry
- DeepSeek
- Groq
- OpenRouter
- NVIDIA
- Ollama
- Opencode
- Custom (OpenAI-compatible)
Transcription only
- Deepgram
- Azure AI Speech
Try the production release
Put the lens on your desktop.
Choose your operating system below; every build is served from GitHub Releases with its SHA-256 checksum attached.
Windowsx64 NSIS installer · v0.7.4Recommended for this device
macOSUnsigned v0.7.4 · choose your chip
macOS · Apple SiliconUnsigned v0.7.4 DMG
macOS · IntelUnsigned v0.7.4 DMG
Linux · x64AppImage · v0.7.4
Linux · arm64AppImage · v0.7.4
Version 0.7.4 publishes the Windows installer, Linux AppImages, and unsigned macOS DMGs. On macOS, clear the quarantine flag with the xattr command in the installation steps; right-click → Open also works on most versions. The macOS build will be signed with an Apple Developer ID and notarized once release signing credentials are configured. Runs on macOS 13 or later, Windows 10 or 11, or a recent Linux distribution (experimental).
What's new in v0.7.4
- Clearing the transcript mid-answer no longer leaves the composer stuck, and a failed stream keeps the text you already received.
- A stalled provider falls through to your fallback instead of reporting a timeout, and long meeting recaps keep the parts they finished.
- Restoring a meeting backup into a full history is refused rather than reporting success after deleting the restored records.
- Exported transcripts use the same clock as their header, and restoring default shortcuts now survives a restart.
- Secondary text clears 4.5:1 over bright slides, readable text stays at 10px or more, and answers are selectable and copyable.
- Provider keys are never written to disk without a working keyring; on a machine without one they last for the session only.
- Local Whisper stays off unless a whisper-server binary is present, instead of offering an engine that cannot start.
Built with boundaries
Useful context still needs judgment.
Capture permissions are explicit, but local consent and recording laws still apply. Do not capture private conversations, protected work, exams, or interviews when you lack permission.
Capture exclusion is best-effort. On macOS the app asks the system to keep its window out of captures; on Windows 10 2004 or later it uses WDA_EXCLUDEFROMCAPTURE. Linux has no capture-exclusion API, so the window can appear in screen shares there, and newer macOS capture paths or a physical camera can still record it anywhere. Never rely on it to bypass disclosure, supervision, proctoring, or security controls.
Source code is available under the Apache License 2.0.
Before you install
A few direct answers.
Does Volyx Lens operate its own AI backend?
No. You configure your own supported provider. Requests travel from the desktop app to that provider, and the provider's own data terms apply.
Where are API keys stored?
The app uses Electron secure storage backed by the operating system (macOS Keychain, Windows DPAPI, Linux libsecret) when available. If secure storage is unavailable, keys are kept in memory for the session only and are never written to disk; Settings says so and lets you update or remove them. They must be entered again after a restart.
Is the release notarized?
The v0.7.4 macOS DMGs attached to the release are unsigned, so macOS blocks a plain double-click. Drag Volyx Lens into Applications, then run xattr -cr "/Applications/Volyx Lens.app" once in Terminal, or right-click the app and choose Open. If macOS says the app is damaged, that Terminal command is what clears it. The macOS build will be signed with an Apple Developer ID and notarized once release signing credentials are configured. The Windows and Linux builds ship unsigned; every file has a .sha256 checksum, a CycloneDX SBOM is attached per platform and architecture, and Windows shows a SmartScreen "Unknown publisher" warning.
What works offline?
AI responses require a configured provider; with a local Ollama server they run fully offline. Realtime transcription uses your configured provider (OpenAI, Azure Foundry, Deepgram, or Azure AI Speech), and Local Whisper covers offline batch transcription only when a whisper-server binary is configured (the published build does not bundle one); automatic cloud fallback remains off by default.
Can Volyx Lens detect and save meetings?
Yes. Meeting detection is opt-in; meeting history is on by default for new installs. While listening, meeting detection looks for a sustained two-sided conversation between the Mic and System channels, then tags the session locally with a "Meeting in progress" indicator. With meeting history on, saved meetings get a meeting badge and stay searchable; you can export records, transcripts, or generated meeting notes. Detection runs in-session. Saved history stays on this device and Volyx Lens never uploads it; generating notes or cleaning a saved transcript does send that transcript to the provider you configured.
Can I choose the transcription language?
Yes. The settings screen lets you pick a language for transcription or leave it on auto-detect. Language selection is honored per job and normalized for both cloud providers and the optional local Whisper provider.
Can I use it commercially?
Yes. The public source is licensed under the Apache License 2.0, which permits commercial use, modification, and redistribution.