Upload Once, Use Everywhere — How GatifyAI Handles Attachments

Every AI chat app can accept a file upload. Almost none of them let you switch models afterward without uploading it again. GatifyAI treats that as a bug worth fixing, not a limitation to live with.

The problem with per-model uploads

Attach a PDF to a conversation with one model, and the moment you switch to another model — whether by choice or because of an automatic fallback — that file is gone from the new model's context unless you upload it again. Multiply that across a long research session with several documents, a codebase, and a couple of screenshots, and "just re-upload it" becomes real, repeated friction.

How GatifyAI does it instead

Attachments in GatifyAI are scoped to the conversation, not to whichever model happened to be active when you uploaded them:

• Upload once. PDFs, images, code files, CSVs, and ZIP archives all attach directly to the chat.
• Every compatible model can read it. Switch from one model to another mid-conversation and the same files are still there, with no re-upload step.
• It survives fallback, too. If GatifyAI automatically switches you to another provider because of a rate limit, your attachments carry over exactly like your message history does.

What's actually stored

Files are kept behind a storage interface that isn't tied to any single disk or provider, which means the underlying storage can evolve without changing how attachments behave from your side. Archives like ZIP files are stored as-is rather than indexed file-by-file — GatifyAI keeps the archive available, it just doesn't currently search inside it.

Why this matters more than it sounds like it should

Attachments are exactly the kind of thing that quietly locks you into one model. If switching providers meant losing your files, every fallback or manual switch would come with a tax you'd feel immediately. By making attachments a property of the conversation instead of the model, GatifyAI keeps the promise of "switch freely" honest — the whole point of routing and fallback falls apart if you lose your context every time it happens.

The takeaway

Upload a document, an image, or a codebase once. Ask your question with GPT, follow up with Claude, and let a fallback move you to DeepSeek if a provider hiccups — the file is there for all three, because it was never actually tied to any of them.