res.mediaset.…, lives in a project like a dataset, and takes the same calls as a file dataset, under /mediasets.
Media types
A mediaset declares one media type at creation, and it never changes:GET /mediasets/media-types lists the registry with the extensions and content types of each format. Every upload is checked against it: a file whose extension is not a format of the media type is refused with 415, and a content type that says otherwise is refused too. A sync applies the same rule file by file, skipping what does not match rather than failing.
Create
PATCH /mediasets/{id} changes the name, slug or description; DELETE removes the mediaset, and with it the sync that fed it.
Files
The files endpoints mirror a dataset’s, with the mediaset’s id in place of the dataset’s.Filling it from a connection
The usual way to fill a mediaset is a file sync from a connection: an SMB share, a bucket, a SharePoint library, a Google Drive, a git repository. The sync copies the files matching its path and glob, keeps those of the media type, and runs on demand or on a schedule like any producer. Building a mediaset runs its sync. A Python transform may also read a mediaset as an input: the build hands it the files directory, and the code walks it.The document pipeline
The platform ships transforms that take a document mediaset from files to vectors, configured with parameters rather than code:
The result is a vector dataset with a producer and a place in the lineage, searched by the agents bound to it. Wire the chain from the Atlas interface or the transforms API; each step takes its settings (OCR engine, chunk size, embedding model) as
parameters.
In an agent
A mediaset bound to an agent gives it three tools:list_files, summary and presigned_url, so the agent can find a document and hand back a link to it. The content itself is reached through the vector dataset’s search. See Agents.
These endpoints are served by the platform and used by the Atlas interface; they are being opened on the public gateway progressively, and the Python library does not wrap them yet. The API reference lists what the gateway exposes today.