To get a large file into Numonic through ChatGPT or Claude, ask the assistant to store it. It gives you a one-time link, you drop the file on that page in your browser, and the assistant checks what arrived and files it for you.
The browser step is the part people do not expect, and it is not a limitation we added. It is where the transfer has to happen. We hit it ourselves on 18 October 2026, when a Claude Code session finished a 62 second video and we asked it to put the result, the source clip, and the pieces into Numonic. That is six files and about 100 MB. The session could read every one of them. It could not send a single one.
Why Can’t the Assistant Upload the File Itself?
A chat assistant works on your files inside a locked-down sandbox. It can read them there. The sandbox has no network access, and the Numonic connector cannot read files from it. That leaves one option for the assistant: paste the file’s bytes into a tool call as text.
That route is capped at about 3 MB per file, because the request body for the connector is limited to 4.5 MB. It also makes the assistant spend minutes re-typing a large file, and long re-encodes often get cancelled. For a video, it is not a real option.
So the assistant cannot upload your large files. What it can do is arrange for you to, and then take over once they have landed.
How Do You Get a Large File Into Numonic Through an Assistant?
- Ask the assistant to store the files in Numonic. It creates an upload link and gives it to you, usually together with the collection it plans to file them in.
- Get the files onto your own computer. If the assistant created them in the chat, download them first (a ZIP is fine, then unzip it). The link takes the files themselves, not an archive.
- Open the link while signed in to the same workspace. Anyone else, or a member of a different workspace, is refused.
- Drop the files on the page, choose Upload, then I’m done. The upload goes from your browser straight to Numonic, so its size is not limited by the assistant.
- Tell the assistant you have uploaded. It reads back what arrived and carries on with filing and checking.
Handing files to Numonic through an assistant
In our run, the whole exchange took four prompts. The assistant created the collection, staged copies of the files under clear names, and produced the link. We dropped the files, and it filed them into sub-collections for the final video, the inputs, and the intermediate pieces.
How Do You Know the Right Files Arrived?
The assistant can pass a SHA-1 checksum for each file it expects when it creates the link. After you upload, it reads back which checksums matched, which are missing, and which were unexpected.
That turns “I think they all went up” into a list. In our run, six files matched on the first drop and the readback named the one that had not arrived. There was nothing to guess at, and nothing to re-upload except that one file.
What Happens After the Upload?
Numonic reads the prompt, workflow and models embedded in each file, such as a ComfyUI graph inside a PNG or MP4, and the asset page shows where that information came from. The assistant then adds the files to the collections you chose. Uploads are attributed to the assistant’s own connection, so you can tell later what it brought in.
This is the part that makes the browser step worth it. The transfer is the only manual step. The lineage, filing and checking all happen without you.
When Do You Not Need the Link?
- Files under about 3 MB. The assistant can store them directly with the connector’s
StoreAssettool, and you do nothing. - Agents that have network access. An agent with a shell and outbound access can request a signed upload URL and send the bytes itself. The link exists for assistants that cannot.
What Are the Limits?
- A link works for one upload session and expires after 30 minutes by default, or 4 hours at most.
- It closes when you choose I’m done or when the expected number of files has arrived.
- It is bound to the workspace and the connection that created it, and you must be signed in as a member of that workspace.
- The assistant needs write permission on its Numonic connection.
The full reference, including the tool inputs, is in the MCP quickstart. For the design thinking behind building a library that agents use as readily as people, agent-first interface design and MCP for creative tools go further.
The assistant cannot carry your files. It can make sure the right ones arrive.
Key Takeaways
- Chat assistants read your files in a sandbox with no network, so they cannot upload large files themselves.
- Ask the assistant to store the files. It makes a one-time link; you drop the files in your browser.
- Checksums let the assistant prove which files arrived and name any that did not.
- After the upload, extraction, filing and checking need nothing more from you.
- Under about 3 MB, none of this is needed.