Files
View as Markdown · Open in Claude Desktop
A tool reads its input from the source field. It takes an https:// URL that returns the file directly, or a file_id from an upload or from an earlier job. A tool that draws from your allowance needs a file_id: the API measures an uploaded file before it reserves time, and refuses a URL with 400 BAD_REQUEST and the fix "Upload the file with create_upload and pass its file_id." A tool that draws nothing, such as probe_media or convert_subtitles, takes either. An address on neattrim.com itself is not accepted as a source: upload the file instead.
Upload a file
POST /v1/uploadswith the file name, size and type. The answer says whether to send one presignedPUTor several parts.- Send the bytes. For a multipart upload,
PUT /v1/uploads/<id>/parts/<n>for each part.GET /v1/uploads/<id>lists the parts the server already has, so an interrupted upload can resume. POST /v1/uploads/<id>/complete. Thefile_idis in the answer of step 1 and can be used once this call succeeds. The quickstart shows the three calls with curl.
The SDKs do these steps for you with upload(). Your plan sets the largest file you can upload: 200 MB on Free, 2 GB on Plus and 4 GB on Pro. A file above the limit gets 413 TOO_BIG.
URLs we refuse
We fetch a URL only when it returns the media file itself. We refuse links to video-platform pages, private network addresses and anything that redirects to one. The answer is 422 SOURCE_REFUSED with the reason.
Outputs and expiry
A finished job lists output files. GET /v1/files/<id> returns the file details and a fresh signed link. Download links made through the API stay valid for 24 hours; ask again for a new one.
Results are kept for 7 days. Your library holds up to 1 GB on Free and 5 GB on paid plans. When the library is full, the oldest items are deleted. Set keep to false on a job to delete its results after 24 hours. DELETE /v1/files/<id> removes a file at once.