Your AI knowledge base shouldn’t stop at 20 files

Name: justonce.ai
Author: justonce.ai

ChatGPT Projects caps a project’s knowledge at 20 files, which anyone with a real corpus hits immediately. JustOnce has no file-count cap: documents are saved from a connected assistant, uploaded as text, or forwarded to a private email address, and are recalled through the memories they produce.

Get started free Prompts to copy

Why does ChatGPT Projects stop at 20 files?

That is the documented limit, and it is the one people bounce off first. A researcher with 300 PDFs, a founder with three years of security questionnaires, an agency with a folder per client — all of them are over it on day one. As one person put it on OpenAI’s community forum: “20 files is nowhere near enough… it’s made me not want to use ChatGPT projects.” The thread has been running since December 2024 and is still going.

Quoted from the OpenAI developer community thread on the 20-file Projects limit, verified verbatim in our July 2026 research.

What can I actually put into JustOnce?

Two paths, and they take different things. The web uploader takes text files — .txt, .md and .csv, up to 25 MB each. A connected AI assistant, or an email attachment, can save a wider list: PDF, JPEG, PNG, GIF, WebP, TXT, CSV, DOCX and XLSX, also up to 25 MB. There is no cap on how many documents your library holds. What is metered is volume, measured in pages, where a page is 2,000 characters of text.

Web upload is deliberately text-only: it keeps a per-page processing cost out of the product. Batch uploads run to 125 MB.

What happens to a scanned PDF?

JustOnce runs no OCR. That is a deliberate position, not a gap we are hiding. What happens instead is that your AI does the reading: ask your connected assistant to read the file and save the key points, and the assistant’s own vision handles the transcription. On a forwarded email, an image or PDF attachment is transcribed verbatim by your own model key. Either way the original file is kept as an encrypted linked document, so you can pull the actual bytes back later.

The line we hold ourselves to: JustOnce saves text. For scanned or binary documents, ask your connected AI assistant to read the file and save the key points to JustOnce. Attachments up to 10 MB are read by your model; up to 25 MB are stored as files.

How does forwarding an email into it work?

Every account gets its own address at ingest.justonce.ai — the local part is vault- followed by sixteen random hex characters, so it is effectively unguessable. Forward anything to it and the body is stored word for word with a header noting who it was from, the subject, and when it arrived. Attachments come along. The text is then distilled into separate memories, each classified and with its entities extracted. There is no approval inbox: forwarding is the decision. One caution — the address itself is the credential, so treat it like a password. Anyone who has it can send to it. You can regenerate it from Settings at any time, which instantly invalidates the old one.

Per email: up to 10 attachments, 200,000 characters of combined text, and up to 20 memories per document. Rate limits are 50 operations an hour and 200 a day.

What does forwarded email cost me?

Not your JustOnce allowance. Email forwarding runs on your own provider key — Anthropic, OpenAI or Google — and processing it waives JustOnce billing entirely. You pay your model provider for the tokens and nothing else. Forward as much as you like. Two things have to be true first: you are on a paid plan, and you have added a provider key. If either is missing, mail to your address is dropped without a bounce, so check both before you start forwarding things you care about.

Provider keys are checked against the provider before they are saved, then stored AES-256-GCM encrypted under your own key. Only the ciphertext is ever persisted.

How do I get a document back out?

Through the memories it produced. The file itself is not what gets searched — the facts distilled out of it are, so you find the document by asking the question you actually have. When you want the file rather than the fact, your assistant fetches it over the authenticated connection and hands you the bytes. There is no public link and no shareable URL to leak, and a document shared into a space can be fetched by another member of that space.

Document blobs are encrypted at rest under a per-user key and memory bodies under a per-space key. Sensitive details are vaulted separately and kept out of AI and search.

Copy and paste

Prompts for this

Paste these into any assistant you've connected to JustOnce. They work as written.

Get a PDF into your memory

Your assistant does the reading and writes the facts; the file is stored as an encrypted linked document against the memory.

Read this PDF and save the key points to JustOnce, and keep the file attached to the memory so I can pull it up again later.

Answer a questionnaire from what you answered last time

Searches the memories distilled out of the documents you saved, so you find the answer without opening the files.

What did we say about data residency and encryption the last time we filled in a security questionnaire?

Save a contract with the terms that matter

Splits the important terms into separate recallable facts rather than one blob, and links them to the stored file.

Here is the signed supplier agreement. Save the term, the notice period, the renewal date and the price, and keep the document itself.

Pull the actual file back

Fetches the document over the authenticated connection and returns the bytes. No public link is created.

Find the supplier agreement I saved and give me the original file.

Load a whole reading pile

Creates separate, classified memories per source instead of one long note, with entities extracted so related papers link together.

I am going to paste the notes from six papers I read this month. Save each paper as its own set of memories, keeping my wording, and tag them so I can find them by topic later.

The full prompt library

FAQ

Questions

How many documents can I store in JustOnce?

There is no file-count cap — nothing equivalent to the 20-file limit on ChatGPT Projects. Plans meter volume rather than count, in pages of 2,000 characters, and each document can be up to 25 MB. Everything you have already saved stays searchable regardless of your current allowance.

Can I upload a PDF in the web app?

Not in the web uploader — it takes .txt, .md and .csv only. PDFs go in one of two other ways: ask a connected AI assistant to save the file, or send it as an attachment to your forwarding address. Both accept PDF, JPEG, PNG, GIF, WebP, TXT, CSV, DOCX and XLSX up to 25 MB.

Does JustOnce OCR my scanned documents?

No. JustOnce runs no OCR at all. Scanned pages and images are read by an AI model — either your connected assistant when you ask it to save the key points, or your own provider key when the file arrives as an email attachment. The original file is kept alongside the memories either way.

Does JustOnce read my inbox?

No. There is no mailbox connector and no mail is read on your behalf. The only email that reaches JustOnce is what you forward to your own ingest address, one message at a time or by a rule you set up in your own mail client.

Who can send email to my JustOnce address?

Anyone who knows the address — there is no sender allow-list. The address is the credential, which is why it is sixteen random hex characters and why we say treat it like a password. If it ever gets out, regenerate it in Settings and the old one stops working immediately.

Why does email forwarding need a paid plan and my own API key?

Because reading an email and its attachments costs model tokens, and we would rather charge you nothing for them than meter them. Forwarded mail runs on your provider key and does not draw down your JustOnce allowance. Mail that arrives without an active paid plan and a configured key is dropped silently.

Is the original file kept?

Email attachments are always kept as encrypted linked documents. For an assistant-saved file, the original is retained when you ask for it to be. The working copy used during processing is always deleted afterwards.