Preview FAQ

Answers for people invited to the PaperLM preview. For the full detail, see Your Gemini API key, the Privacy Policy and the Terms of Service.

Your Gemini API key

Why do I need my own key? The preview runs every analysis, chat and workspace job on your own Google Gemini API key. There is no shared key on this plan, and no setting can turn one on.

Which key should I use? One from a Google Cloud project with billing enabled (Google's paid tier), ideally a project you use only for PaperLM, with a budget alert. Google's terms require the paid tier for people in the European Economic Area, Switzerland and the United Kingdom, and on the free tier Google may use what you send to improve its products.

What does it cost? Whatever Google bills for the calls your analyses make. Typical figures at Google's list prices with the current defaults: Gemini 4 Argon writes the review, the roast and the translation, Gemini 3.8 Flash does fact extraction and the posts, Gemini 3.5 Flash-Lite finds the figures, and Nano Banana 2.1 draws. The Argon and Nano Banana 2.1 figures are estimates from the measured costs of the previous models scaled by list price, not yet measured:

PartStandardWith Flex
Review (extraction, figures, review, critique, refinement)about $0.90about half
Roast, with the review updated from itabout $1.00about half
Cover imageabout $0.05about half
Comicabout $0.12about half
Social postsabout $0.04about half

So a review with a roast is around $1.90, and everything switched on around $2.10, on the standard tier. A long paper, a bigger model or a 4K image costs more. The Usage tab of each paper, and Settings, list every step with its model, calls and an estimated cost. Your Google bill is the real number.

Can I pay less? Yes: turn on Flex in Settings. Google's Flex tier runs the same models at roughly half the price in exchange for waiting: a stage can take anywhere from a minute to about fifteen instead of starting at once, so a paper takes longer to finish. Chat always uses the standard tier. If Flex has no capacity, PaperLM falls back to the standard tier so the run still completes.

How is the key stored? Encrypted with Google Cloud KMS, bound to your account, in a separate database that has no backups. Only the two PaperLM services can decrypt it, every decryption is audit logged, and an alert watches for a decryption by anyone else.

Is it checked? Yes. When you save it, we make one free call to Google to confirm it works, and tell you in plain words if it doesn't.

Can anyone at PaperLM see it? Only its last four characters, so you can tell which key is saved. No page, export or admin view returns the key.

How do I remove it? Settings → Gemini API key → Remove. That deletes it at once and stops your queued and running analyses (a running one stops at its next stage). If you think the key leaked, also delete it in Google AI Studio: that is what stops spending.

Limits

  • 2 analyses at a time. A third is refused with a message; it is not queued.
  • 50 runs a day, reset at 00:00 UTC. Retries, stage reruns and workspace jobs count. Chat does not; it has its own rate limit.
  • If you need more for a real reason, ask. Limits can be raised per person.

What is recorded, and what isn't

Recorded:

  • Usage events, for example that you started an analysis, that it finished or failed and at which stage, that you viewed a review, or that a limit stopped you. Kept 90 days.
  • A usage ledger: for each model call, which model, how many tokens and the estimated cost. Kept 400 days.
  • Problem reports and feedback you send, with the paper's state if you leave that on. Kept 400 days.
  • Server request logs, including your IP address. Kept 30 days.

Not recorded: your key anywhere outside its own database; the text of your papers, reviews or chats in the usage events, the ledger or our metrics; analytics or advertising cookies.

Where your paper goes: to Google's Gemini API, with your key. The review, critique and roast stages let Gemini search the web with Google Search. Chat does too, and reads pages you link, unless you switch that off in the chat.

Where it is stored: the database, servers and key encryption are in Google Cloud's Belgium region. Uploaded PDFs and generated images are stored in Google Cloud's Iowa region (USA).

Signing up

  • You need an invite code and must be 18 or older.
  • Sign in with Google, or with an email and password. With a password, we email you a link to confirm the address first. The email comes from [email protected]; its link points to a firebaseapp.com address, which is ours.

Your data

Settings → Your data: download everything as one JSON file, or delete your account. Deleting removes your key at once and everything else after 14 days; you can cancel until then, but the key is not restored.

Known limitations

  • It's a model's reading of the paper. It can be wrong or miss things. Check what matters against the paper itself.
  • Minutes, not seconds. An analysis fills in stage by stage.
  • Paywalled papers can't be fetched. Upload the PDF instead, up to 30 MB.
  • Figures are shown only when they can be matched to their caption.
  • Provider trouble fails a stage. When Gemini is overloaded or your key runs out of quota, a stage can fail after its retries. Retry the paper or rerun the stage.
  • Deleting your key mid-run stops a paper analysis at its next stage. A workspace job already running finishes first (at most 30 minutes), and a Deep Research request already sent runs on at Google.

Reporting a problem

Use the ladybug button in the app. It sends your message with the state of the paper you are on (its status, stage and error), never the paper's text, its review or your chats.