The context workbench
A useful detail
changes the draft.
The task stays the same: "Write an invitation." Give it a few useful details and see what each one contributes.
Give the task something to work with.
Tap a card to add it; tap again to remove it. On a desktop, you can also drag cards between the shelf and tray.
The detail shelf
Your invitation
No details yet. The brackets show what is missing.
A highlighted line changed with your last action.
Rule-based illustration, not a live LLM. All event details are fictional example data.
Useful context gives an answer direction.
An audience, a date, or a constraint can change what belongs in the answer. Remove a detail and the draft loses that information. The model also brings learned patterns to a task; supplied context helps it apply those patterns to your situation.
See how the illustration works
Each card controls a part of a prepared invitation. The draft is assembled from explicit rules, so the effect of adding and removing information is clear and repeatable. No language model is being called.
Real LLMs can respond less predictably, infer unstated details, or overlook information. The brackets here intentionally make missing information visible. Adding context during a normal response does not retrain the model's parameters.
Try adding only the tone card. Then remove it and add only the date. Which change gives you a more usable invitation?