Is it safe to let it use my browser?

You watch every step as it happens, and you can stop it at any moment with one click. The instant you use the page yourself, it pauses and waits for you.

Do I need to know anything technical?

No. You describe what you want in plain words — “fill this form”, “compare these three pages” — and it figures out the rest.

What does the free plan include?

Bronze gives you the browser sidekick itself — the agent on your current tab, web search, projects, chats, notes and sync between the extension and the web app — running on the Standard AI, with one usage allowance per 5-hour window. No card required. Four things are not in Bronze: the Smarter AI, downloadable file deliverables, Guidelines, and Libraries. Those start at Silver.

It is your own documents, kept where your AI can read them. Upload contracts, specs, research, spreadsheets or notes; StickyFriend reads each file, indexes it, and from then on any chat you point at that library can search it and quote from it, telling you which file the answer came from. Silver gets 3 libraries, Gold 15, Diamond 50; Bronze has no access.

Can it read scanned documents and images? (OCR)

On Gold and Diamond, yes — those plans run OCR, so scanned PDFs, photographs of pages and plain images are read as well as ordinary files. Slide decks never needed OCR: a .pptx is read as text on every library plan. Silver is text extraction only and refuses a file with no text layer rather than adding an empty document.

What if I hit the usage limit?

You wait for the reset. The window is fixed, not rolling: it starts the moment you first use the Standard AI and runs for five hours, and the whole allowance comes back at once when it ends. Paid plans multiply the size of the allowance rather than shortening the window.

How do golden credits actually work?

Golden credits are the currency for the Smarter AI, and they are not one-per-reply — a credit tracks the work in the reply. Plan credits arrive monthly with a small daily top-up and roll over up to two months' worth. Credits bought in a pack sit on top, do not expire and are spent only after plan credits run out.

Do the pricier plans answer faster?

On the Standard AI, no — every plan is answered by the same self-hosted model and there is no fast lane. On the Smarter AI, Gold and Diamond send their requests to the provider flagged priority while Silver sends them as standard. Gold and Diamond share that same queue.

Can I use it away from my browser?

Yes — the web app has all your same conversations and screenshots, readable from anywhere. Actions on live pages happen in the browser, and it will point you there when needed.

From a web page to a chart you can trust

August 1, 2026

A number you cannot check is just a rumour

Every research task ends in the same place: a figure sitting on a page, and no clean way to get it into a form you can work with. You retype it into a spreadsheet, lose the source somewhere in the process, and three weeks later cannot remember whether it was a total or an average.

The gap worth closing is not “find the number”. It is everything between finding it and trusting it — and that stretch is where research quietly goes wrong.

Step one: take the table, not a picture of it

Ask for the table on the page in front of you and it comes off as data — rows, columns, types — rather than as an image or a paraphrase in prose. The source URL travels with it, so the figure can be traced back to where it was published without anybody having to remember.

This counts for more on the pages that are hard to reach than the ones that are easy. A statistics portal that paginates its results, a catalogue hidden behind three filter dropdowns, an archive whose own search box is the only way in — the assistant works those controls for you, so you arrive at the table instead of clicking towards it for twenty minutes.

Step two: reshape it where reshaping is safe

Real tables arrive wrong. Numbers stored as text. A header row repeated every twenty lines. One column holding two facts. Four files that should have been one. Cleaning that up by hand is exactly where the mistakes get in, because every fix is a small opportunity to fix it slightly incorrectly.

The transformations run in a sandboxed Python environment: filtering, grouping, joining, pivoting, deriving new columns, folding several sources into one table. It is isolated, it touches only your material, and it exists for the length of a single request.

Why a sandbox rather than prose

A language model asked to add up two thousand numbers will produce something that looks exactly like a total. Running the sum in Python produces a total. The difference is invisible in the output and decisive in the meeting where somebody checks it.

Step three: see it

Ask for the chart you actually want — a trend over time, a comparison across categories, a distribution — and it comes back rendered from your figures, with the axes labelled. Ask for a different cut and it redraws, rather than starting the analysis over from the beginning.

Step four: hand it over

When the work has to leave your desk, the same material becomes a styled Excel workbook you can keep working in, a full report, or a slide deck. These are versioned: asking for a change edits the file rather than regenerating it from nothing, so the fifth revision still carries the first revision’s corrections.

Keeping the trail

Every source page you used stays attached to the project with its link, so the citation is written before you need it. When a figure is challenged a month later, the path from the chart back to the published page is still intact — which is the difference between defending a number and re-doing the work.

Extract — tables off live pages and out of your uploaded files.
Transform — clean, join, pivot and derive, computed in the sandbox.
Chart — drawn from your figures, redrawn on request.
Deliver — styled workbook, report or deck, edited in place afterwards.

Four steps, one panel, and the sources still attached at the end of it.