Home AI Compass Basics
Context window
How much text a model can take into account at once — its working memory for a single task.
Basics path · 4 of 7
What it is
The context window is the maximum number of tokens a model can consider in one go: instructions, conversation and documents together. Anything beyond it is cut off, or has to be summarized or retrieved selectively.
Example
A model may handle a long report in one pass, but a whole document archive will not fit — that is what retrieval (RAG) is for.
What can go wrong
Stuffing everything into the context raises cost, slows answers and can bury what matters. A longer context also gives hidden instructions more room.
How to secure it
Send only what the task needs, keep documents of different users and clients strictly separate, and log what was sent.
In Obsigen
In Obsigen the model receives only what the task requires; your data stays on dedicated infrastructure. About Obsigen AI
Related
Related service: Applied AI