Guide
Giving AI the right context, and only the right context
The quality of an AI answer depends heavily on what the model was given to work with. Too little and it guesses. Too much and the relevant part gets lost among the rest. Managing context is most of the skill of working with these tools.
What counts as context
Context is everything the model can see when it answers: your prompt, the earlier messages, and any material attached to the conversation. In Bonscape that material can be a document, a spreadsheet, an image, a link to a web page, or a note you wrote yourself.
- Documents: PDF, Word, PowerPoint and Excel files.
- Text and data: plain text, Markdown and CSV.
- Images: PNG, JPEG and WebP, read for their content.
- Links, notes and sticky notes created on the canvas.
Why “upload everything” does not work
It is tempting to attach every file to one chat and ask questions. The trouble is that the model then weighs all of it for every answer. A question about the contract is answered with the marketing deck in view. Answers get vaguer, and every prompt costs more because all of that material is sent each time.
Connect context where it is needed
On a canvas, a file is a node, and it only informs the chats it is connected to. One requirements document can feed five branches, while a confidential file is connected to a single branch and stays out of the others. You add material once and decide, per conversation, whether it applies.
Because context belongs to the workspace and not to a provider, it also stays put when you change models. There is nothing to re-upload.
Where this matters most
Work that draws on many sources benefits most: a literature review where each paper must stay distinct, or product discovery where interview notes need to stay attached to the conclusions drawn from them.
