Context
The information supplied to the model for the current inference request, including any irrelevant or conflicting material.
Illustrative conversation
- The answer is in our docs, so why did the agent miss it?
- The document existed, but it was not loaded into the model's current context.
Definition
Context is the set of tokens available to the model during inference. It can include system instructions, the user's request, message history, tool definitions and results, MCP content, and retrieved files. Information may exist in a repository or database without being in context; the harness has to load it before the model can reason from it.
Reviewed