Organize context around a project
Keep related source material together in a project with folders and readable filenames. Separate projects give different teams or engagements their own membership and assistant grants.
Shared source material for AI
Bring project briefs, reference notes, research and data files into one place. Give your team and connected assistants permissioned access to the source material they need for the work.
Original files. Project permissions. Access you control.
BUILT AROUND YOUR FILES
Keep related source material together in a project with folders and readable filenames. Separate projects give different teams or engagements their own membership and assistant grants.
Publish a new revision when a brief, reference file or data export changes. Earlier revisions remain in version history, and current-version checks prevent an assistant from silently replacing someone else's newer work.
Assistants can find files by name and read supported extracted text in bounded excerpts. Initial extraction covers UTF-8 text, Markdown, CSV and JSON within processing limits. Other originals remain available for authorized download.
Invite collaborators with project roles and authorize each assistant connection separately. Removing membership or revoking a connection blocks subsequent requests through that access path.
HOW IT WORKS
Start with a focused set of reference notes, current requirements or research files. Use clear names so people and assistants can identify the relevant material.
Add the files to a project, organize them into folders, and assign roles to collaborators. Review which projects each connected assistant is allowed to access.
Ask a connected assistant to retrieve the relevant files for its task. When source material changes, upload a new version and use the returned version information to identify the revision being used.
A FEW PRACTICAL ANSWERS
It contains the source material you want an assistant to use: reference notes, briefs, instructions, research and data exports, for example. In Files for AI, those sources remain ordinary files with project permissions and version history.
Uploading stores the file in your project. It does not start a model-training job. When you authorize an external assistant to retrieve material, that client's data handling and its provider's terms also apply.
No. Filename and metadata search are available across stored types. Content search applies only after supported text extraction succeeds within its limits. Unsupported, failed and over-limit extraction remain visible states; the original file stays downloadable.
The connected assistant produces its response using the material it retrieves. Files for AI provides source storage, access and retrieval; it does not currently provide a built-in assistant conversation or semantic search service.
FILES FOR AI
Bring your files together and choose how your assistants can use them.