Natively-AI-assistant/natively-cluely-ai-assistant v3.2.0

Natively v3.2.0

on GitHub

Summary

Ready from the first minute — a guided first run, smarter Auto Answer, faster screenshot answers, Answer length, every answer on your phone, and plans by the hour.

What's New

• First-run guide — new users walk through two short meetings, one in the overlay and one in the notepad, then their notes, plans and keys. The first meeting plays a demo film to practise on. Show it again from Settings → Setup & Help.
• Smarter Auto Answer — an on-device model decides which questions are meant for you and answers only those. Its settings are in Settings → Intelligence; the switch stays in General. Part of Pro, or free when you share the transcripts of the meetings where it runs.
• Faster screenshot answers — the first word of a screenshot answer arrives in about 4 seconds instead of about 10. Choose the question a screenshot asks when you type nothing in Settings → Overlay → Screenshot question.
• Answer length — Shorter, Smart or Longer answers, one setting in Settings → Overlay for every mode. Shorter and Longer are Pro.
• Every answer on your phone — each overlay answer (buttons, Auto Answer, Direct Assist and spoken follow-ups) streams word by word to your paired phone.
• Newer speech models — Soniox v6 and ElevenLabs Scribe v3 with your own key, and Nepali in the language picker where the provider supports it.
• Plans by the month, the year or the hour — Natively plans can be paid yearly with two months free, and one-time Light ($5, 2 hours within 7 days) and Heavy ($10, 5 hours within 14 days) packs include Pro. Plans & Billing has an Hourly, Monthly and Yearly switch.
• Lost your key — "Bought a key but no email?" sends your key again, from the activation card in Plans & Billing.

Improvements

• Auto Answer — waits for announced lists and scenarios, answers back-to-back questions in one answer, and never answers a bare transition.
• Live answers — follow the Real-time prompt, skills and Cue Cards on every path, show the source badge their evidence earns, and read a follow-up against the previous answer.
• Speed — the answer path warms up when the meeting starts, small talk gets no early answer, and more of each prompt is reused from the cache.
• Screen memory — screenshots are remembered in typed chat, What to answer, Code hint, Brainstorm, your own buttons and phone questions.
• Keep overlay size — the overlay keeps its size between answers. On by default, in Settings → General → Advanced.
• Search — questions you hear also search your files with the on-device embedder.
• Banners and cards — one full-window card per launch, calmer meeting banners, and the auto-stop warning shows above the rest.
• Calendar — reminders refresh themselves and fire only when they should, with one offer per call.
• Packs — Natively says when a pack has ended, and the meeting resumes after a top-up.
• Open at login — on by default for fresh installs.
• Look — a lavender accent with periwinkle switches, every Settings title in Inter Semibold, and darker Settings cards in dark mode.

Fixes

• Overlay — keeps your place when an answer you didn't ask for arrives (#723).
• Coding questions — "show it in python" keeps the right problem, and "Java 3" is no longer read as a language rule (#737, #738, #739, #740).
• System audio — capture restarts after a failed start (#729). On Windows it falls back to the communications output when loopback can't use the default device.
• Microphone — speaker sound leaking into the microphone no longer counts as you.
• Speech provider — no silent Google fallback when your chosen provider has no key.
• Natively speech — keeps the sentence spoken just before Stop.
• Local speech models — switch without a restart, and a local model's rewritten partials no longer stay on the transcript bar.
• Timeouts — long Natively answers get 30 seconds instead of stopping at 8.
• Provider errors — a failed provider chain says the providers failed instead of "none set up".
• Typed questions — a question typed while an answer streams is sent instead of ignored.
• Screenshot question — the field keeps its spaces and reaches the overlay.
• Interview answers — a design problem stated over several sentences gets a design answer, and a screenshot with a non-coding question is no longer answered as code (#741, #742).
• Follow-ups — an answer cut off part-way is no longer used as context for the next one (#743).
• Meeting notes — figures such as "40M deliveries" reach the notes as counts, while salary and pricing lines stay out (#744).

Technical

• Auto Answer's on-device model is downloaded in Settings → Intelligence, not shipped in the installer.
• Auto Answer lives in the premium module; the source-available build shows no Auto Answer rows.
• Telemetry records how the previous run ended (app_run_ended), to find crashes.
• Electron 43.7.7 and Node 22.13 or later, unchanged from 3.1.0.

macOS Installation (Signed Build)

Download the correct architecture .dmg or .zip file for your device (Apple Silicon or Intel).

• For Apple Silicon (M1/M2/M3/M4): Download the arm64 build.
• For Intel Macs: Download the x64 build.

Open the downloaded file, drag Natively to your Applications folder, and launch it.

⚠️Windows Installation (Unsigned Build)

When running the installer on Windows, you might see a "Windows protected your PC" warning from Microsoft Defender SmartScreen saying it prevented an unrecognized app from starting.

Since this is an unsigned build, this is expected. You can safely ignore it by clicking More info and then Run anyway.

See CHANGELOG.md for the detailed list of changes.