Introducing Persistent Files: A Permanent, Searchable Library for Your AI
You found the perfect document — the contract, the research PDF, the style guide — and handed it to your AI. It was brilliant. Then the chat ended, and the file went with it. Next conversation? Upload it again. And again.
Persistent Files end that loop. Upload a document once, and every one of your ARMES agents can find it, summarize it, and read exactly the parts that matter — in any chat, anytime. Think of it as giving your whole AI team a shared, permanent library.
The problem with attaching a file to one chat
The classic way to give an AI a file is to attach it to a single conversation. It works, but it has two quiet costs.
First, the file is trapped in that one chat. Want it in tomorrow's conversation? Re-upload it. Your knowledge never accumulates — it evaporates every time a thread ends.
Second, attaching dumps the entire file into the conversation. Fine for a one-pager. Terrible for a 90-page report, which floods the chat, crowds out everything else, and burns through the conversation's working memory before you've even asked your question.
Persistent Files solve both — durable reuse, and a smarter way to handle big documents.
Meet Persistent Files
Upload a file once; every agent can search it, summarize it, and read it — in any chat, anytime — instead of you re-uploading it over and over.
Drop in a contract, a spec, a spreadsheet, a style guide, a stack of research papers. From that moment on, the file lives in your library alongside your notes, gets automatically summarized, becomes searchable by all your agents, and can be pulled into any conversation on demand.
Chat attachments vs. your library
Both still exist, and both are useful. The difference is lifetime and reuse.
| Chat attachment | Persistent file (your library) | |
|---|---|---|
| How you add it | Paperclip / "Add Attachment" in the chat box | "Upload File" or drag-and-drop into your library |
| Lives in | Just that one chat | Your library, organized in folders |
| Lifetime | Temporary (expires after ~10 days) | Permanent (until you delete it) |
| Reuse across chats | No — re-attach each time | Yes — available in every chat |
| Searchable by your AI | Only within that chat | Yes — across all your chats |
| Can be scoped to a Project | No | Yes |
The rule of thumb: a quick one-off goes on the chat; anything you'll reference again belongs in your library. And if you attached something in chat and realize it's a keeper, you can promote it to your permanent library with a single click — no re-upload.
How a file becomes AI-ready
When you add a file, ARMES prepares it automatically:
- Upload — add a file to a folder via the button, drag-and-drop, or "Save to Library."
- Read & summarize — ARMES extracts the text and writes a concise AI summary of the document.
- Ready — the summary and full text become searchable and readable by every one of your agents.
You can watch it happen. Each file shows a live status, and the app keeps it fresh on its own:
| Status | What it means |
|---|---|
| Pending | Queued |
| Processing | Being read — summary and text are on the way |
| Ready | Done, fully usable |
| Failed | Something went wrong — a "Retry processing" button is right there |
Supported files: up to 8 MB each, including images, PDF, Word/Excel/PowerPoint, plain text, CSV, HTML, XML, YAML, JSON, and OpenDocument formats.
Big documents, handled gracefully
This is where Persistent Files quietly shine, and it reflects a genuine best practice in AI design: summarize first, retrieve on demand.
Old-style attachments shoved the whole document into the conversation at once. Persistent Files take a smarter path:
- Every file gets a concise summary up front — that's what your AI sees first, so it instantly knows what the document is about.
- When specifics matter, your AI reads deeper — the full text, or a precise section at a time — pulling in only the parts your question actually needs.
- Even very large documents are read in manageable pieces rather than all at once.
Why does this matter beyond convenience? Every AI conversation has a finite working memory (its "context window"). Stuffing a hundred-page PDF into it wastes that space and dilutes the model's attention. Fetching only the relevant slice keeps the AI focused, fast, and accurate — and lets you keep enormous reference documents in your library without ever overwhelming a chat. It's the same principle behind the most reliable AI systems, built right into how files work.
How you bring a file into a conversation
There are three easy paths:
- Mention it. Type
@in the chat box to search your notes and files, and pick the exact file you want. It's the best way to say "use this specific document." - Attach it for a one-off. The classic paperclip flow still works for something you only need in this one chat.
- Promote a chat attachment. Sent a file and realized you'll want it again? Hit Save to Library, give it a title and folder, and it becomes a permanent, searchable file — no re-upload.
In every case, your AI starts with the summary and digs into the details only as needed.
Real ways people use Persistent Files
- A reusable SOP or playbook — upload your standard operating procedure once; any agent in any chat can answer from it. One source of truth for your whole AI team.
- A long contract or spec — drop in 60 pages and ask about specific clauses; your AI reads the exact sections without choking on the whole thing.
- A research desk — build a small library of papers and reports, then ask your AI to compare them or pull a stat.
- A brand or style guide — keep your voice and editorial rules on file so every writing agent stays on-brand, without you re-explaining the rules each time.
- Spreadsheets and datasets — upload a CSV and have your AI read just the parts that matter, instead of dumping thousands of rows into a chat.
Works beautifully with Projects
Persistent Files honor your Projects, too. Scope a file to a specific Project to keep client, work, or personal material separate — while shared reference files left in "All Projects" stay available everywhere. Your AI sees the active Project's files plus your shared ones, and nothing from other Projects. (New here? Read our companion post on Projects.)
On every plan — including Free
Persistent Files work on the Free tier, on Pro, and on Ultra. A permanent, searchable knowledge base for your AI isn't a luxury add-on — it's how AI should have worked all along.
The best way to feel it: upload one document you keep reaching for, then start a fresh chat and ask your AI about it. It'll already know.
Or start free — no credit card required. Same library, same permanence, on the house.
Upload once. Use everywhere. Forever.
Written by
ARMES Team
From the team building ARMES — private AI that puts every frontier model in one place.