The actor anchors a request in specific material — a file, a selection, a source, a live connection — so the system works from the actor’s ground rather than its generalities.
Forms
Recurring forms: file and image attachment, @-mentions that pull a named thing into the request, removable context chips showing what is in scope, and standing connectors to external sources.
Resources & references
- Shape of AI — Attachments, Connectors
- aiuxpatterns.com — Reference Material
- AI UX Playground — Context Chip Management, Context Mentions, Multimodal Input
To-do
Prompt hosts an @-mentions stub of this territory; migrate it here as this matures. The hand-attached / auto-retrieved boundary deserves a forces treatment.
Consequences
- the output is anchored in named material, and can point back to it — the attachment that went in is what a citation can later point out
- the attached material scopes the answer, so wrong, stale, or partial material scopes it wrong with the same confidence
- supplying context grows into its own task: what began as attaching a file becomes curating a corpus the system works from
Related patterns
Precedes
- Citation — the output is anchored in named material, and can point back to it — the attachment that went in is what a citation can later point out
Enables
- Prompt — grounding material is a constituent of the articulated request
Related
- Data & Information — standing connections to sources sit at the material layer the data foundation describes
- Searching — retrieval does implicitly what attachment does by hand — choosing what the system works from