Context window
The finite amount of information a model can consider in a single provider request.
Illustrative conversation
- Can we paste the entire monorepo into one prompt?
- Only if it fits, and focused retrieval usually gives the model a cleaner context than sending everything.
Definition
A context window is the maximum token capacity available to a model for one request. Instructions, history, tool definitions, and retrieved evidence consume capacity. Output and reasoning accounting depend on the model and API, so leave the documented headroom. A larger window does not retrieve files or guarantee that every supplied detail will be used correctly.
Reviewed