Why do I have to upload the same document to every AI?
A file you upload lives inside the assistant you uploaded it to, and the next one can't see it. Here's what actually happens to a long document, what "personal RAG" means, and when uploading it to the chat is still the right move.
The short answer
An uploaded file stays where you put it. Each assistant keeps uploads inside its own service, and they don't read each other's. Add a report to one, and as far as the next one is concerned it was never added at all.
So the document follows the tool rather than the work. Switch assistants, or add a coding tool to the mix, and there's another place that wants the same file.
What actually happens when you upload a long document
An assistant does one of two things with it: reads the whole thing into the conversation, or — when it is too large for that — looks up only the parts it needs.
When the whole document is read in, its full length arrives again in every new conversation. Nothing carries over from the last time you asked. Come back to it on three different days and it is read in three times. That length is also what you pay for, which is easiest to notice in a coding tool, where a working day is a series of fresh sessions rather than one long chat.
When the assistant switches to looking things up, it stops seeing the document as a whole. It fetches the passages that match what you asked and works from those. The switch happens on the assistant's own terms, and you are not told which mode you are in.
Neither is the wrong behavior — they are two reasonable answers to a document that doesn't fit. What matters is what follows from the second one: past a certain size, looking things up is already what's happening. The open question was never whether to use retrieval. It is whose retrieval it is, and whether it follows you.
Is "personal RAG" the answer?
It is a description of what is already running, not a project you have to start.
RAG — retrieval-augmented generation — means finding the relevant text first and having the model write its answer from that, rather than from what it already holds. When the text being found is your own documents, people call it a personal RAG.
You can build one yourself, and there is a well-worn path for it if running the whole thing is what you want. But building it answers a different question than the one that started this. The question was never how retrieval works. It is where the document lives — because that is what decides whether the next assistant can use it.
The part that has to live outside the assistant
For one document to be usable in any assistant, it has to sit somewhere that belongs to none of them. A store like that is called a knowledge layer — the same idea that applies to notes and decisions, applied to files.
What changes for a document is the number of copies. You add it once. After that, every assistant you connect looks things up in the same place, and changing tools doesn't mean adding the file again.
When you should just upload it to the chat
If you're going to use the document once, upload it to the chat and move on. Putting it somewhere permanent is work you won't get back.
- One-off analysis. "Pull the risky clauses out of this contract." You want an answer about this file today, and then you're done with it.
- Anything computational. Totals across a spreadsheet, a chart, a reformat — that is a job for the tools that run code on the file, not for a store.
- Going deep on one set of sources. Reading a stack of papers together with everything cross-referenced is what study tools are built for. Different job, different shape.
The case for keeping a document outside shows up later, when you notice the same file going up a second and a third time.
What this looks like in practice
You add the document once, and after that you just ask. The file stops being something you carry between tools and becomes something the tools reach into.
Contextick is one of these stores. You upload a file from your dashboard, it is converted to text and saved, and any assistant you've connected can find it later — here's how uploading works.
FAQ
If I upload a file to ChatGPT, can Claude read it too?
No. Each assistant keeps uploads inside its own service and they don't read each other's, so the file has to be added to each one separately — or kept somewhere outside both.
What is retrieval-augmented generation?
Finding the relevant text first and having the model write its answer from that, rather than from what it already holds. When the text being found is your own documents, people call it a personal RAG.
Why does a long document cost more every time I use it?
Because when it is read into the conversation, its full length arrives again in each new one. Nothing carries over from the last time you asked, so the same document is read in as many times as you come back to it.
Where should a document live if I want any AI to use it?
Somewhere that isn't inside any single assistant. A store that sits outside them can be connected to each one, so the document is added once rather than once per tool.