Guide

Media library quickstart — upload a PDF, extract its text, let an agent read it

Five minutes from a PDF on your laptop to an agent answering questions about it — through the project's media library, no scripts.

9/13/2026 · connect0 · 5 min

The media library is where files go before and after agent work. This guide does the most common thing with it: a document in, an agent answer out.

1. Open the library

In the dashboard pick your account and project, then Media in the sidebar — or go straight to /a/<slug>/p/<project>/media. The grid is empty on a new project; the Upload button is top right.

2. Upload the PDF

Click Upload (or drop the file anywhere on the grid). The tile shows up immediately as pending and flips to ready a moment later — the bytes go from your browser straight to storage on a short-lived signed link, so a 40 MB annual report takes as long as your connection, not a server's.

Documents are capped at 50 MB, media at 500 MB; anything larger is refused at upload with the cap in the message.

3. Extract the text

Select the tile and click Extract text. A background job produces a second artifact — the same document as Markdown, marked derived and linked to the PDF — which is what an agent can actually read. Large PDFs take a minute; the tile resolves in place, no reload needed.

4. Send it to the workspace

Select the derived Markdown artifact and click Send to workspace. It lands at /workspace/media/<filename>.md in the project's workspace: the working directory every agent on the project sees.

5. Ask an agent

Open an agent on the project (or create one) and ask:

Read /workspace/media/annual-report.md and list the three largest line items with their year-over-year change.

The agent reads the file with sh (cat /workspace/media/…) and answers from it. Nothing about the file was pasted into a prompt by hand, and the next agent on the project can read the same file.

Going the other way

When an agent writes a result to its workspace — a CSV it built, a chart it rendered — type the path into Add from workspace on the Media tab (data/summary.csv) and it becomes a library artifact you can download or keep. The library is the durable, human-visible side; the workspace is the agents' scratch space; crossing between them is always an explicit step.

What it costs

Storage meters per the rate on pricing; text extraction is a document job at the published document rate. Both show up on the usage page as their own lines — see Credits and pricing.