Common questions

No. Your spreadsheet contents, column and dataset names, and anything you type into chat never leave your machine. This is a hard boundary in how CLAIR works, not a policy promise. See the network contract for exactly what does cross the network.

A short, named list: signing in and renewing your license, checking for app updates, one-time model and library downloads, and optional telemetry. Full detail on the network contract page.

Yes, during the alpha, so we can find bugs and improve performance. It's anonymous and content-free: no spreadsheet contents, dataset names or chat text is ever included. You can see exactly what's been collected and switch it off at any time in Settings, and collection stops immediately when you do. See telemetry.

During the alpha, access is assigned by the CLAIR team to an invited cohort. If you weren't already invited, use the request access form to ask. Once invited, you sign in at the gated portal with your invited email. Full walkthrough on the download page.

No. The Windows installer installs just for your own user account, with no administrator rights required. It works fine on a university-managed laptop.

You'll sign in with a 6-digit emailed code, and CLAIR downloads its AI model and a one-time library pack in the background, a couple of minutes on most connections. After that, it works offline. Details on the download page.

Not yet. The alpha ships Windows 10/11 only. A macOS build is planned, but packaging and notarization work for it hasn't started.

Reach out any time, with a bug, a question, or a feature idea. You'll reach a real person, not a ticket queue, and every message genuinely shapes what gets built next. See contact.

The Deep Analysis tier runs a genuine, staged AutoML pipeline: Bayesian hyperparameter search across several model families, cross-validated feature selection, out-of-fold stacking, and a champion/challenger lifecycle, checkpointed so a run can span multiple idle windows. See Deep Analysis.

An answer is marked verified only when it comes from deterministic SQL, a persisted model, or a rule-based computation. A guessed or hallucinated number is never stamped verified, and this is covered by a dedicated test suite. See Ask.

Yes. It remembers where the file lives on disk, detects real changes cheaply, and refresh shows you exactly what moved (row and column counts) with a restorable snapshot, so you can always see what changed and undo it. See Files and refresh.

Yes: high-resolution figures with captions, a narrative report stitching your findings into one PDF or PowerPoint, chat transcripts that include the underlying rows, and dashboard boards that each export as a single file. See Exports.

Almost certainly, at some quality tier. The base tier has no RAM floor at all; larger tiers become available at 8, 12, 16 and 32 GB of RAM. Full breakdown on the hardware page.

No. Every tier runs on CPU alone, just more slowly. If you do have an NVIDIA, AMD, or Intel Arc GPU, CLAIR detects and uses it automatically, with nothing to configure. See hardware.

CLAIR estimates how long a tier will take to answer before recommending it, so you're not defaulted onto a tier that would feel broken. If any single answer does take more than 20 seconds, it offers a one-click switch down a tier, naming the real measured time.

Free during the alpha. No card, and access is assigned rather than purchased. See pricing.

Yes. Activating CLAIR binds it to the specific machine you're using. Moving to a new device rebinds it there, with a 30-day cooldown between rebinds.

No. Signing in and renewing happen online; using CLAIR day to day is entirely offline. It checks in quietly in the background, and a spotty connection never locks you out mid-session.

One click, in-app. When a fix or improvement ships, CLAIR shows an "Update & Restart" banner. It is never forced; you can keep working and update whenever suits you.

Pricing beyond the alpha hasn't been set yet, and this page will say so plainly with real numbers once it is. Nothing converts to a paid plan without you being told first. See pricing.