Contextual Knowledge
Pengetahuan Kontekstual
Facts the agent can read directly from the context right now. Counterpart to parametric knowledge.
Contextual knowledge is what the agent can read directly from the context right now: the type definition in the file it just opened, the decision written in the spec, the value returned by the last tool call. It is always correct and always current, because it is a copy of the thing itself.
The counterpart is parametric knowledge, what the model knows from training. That is frozen at training time and can be outdated, which is why the two are treated differently when a system is designed. Contextual knowledge can be trusted without a check; parametric knowledge should be checked, and when the two conflict, the contextual one wins because it came from the source.
The practical test for a good agent system is how quickly it can turn a question into contextual knowledge. A model that knows the answer only in its parameters will answer from memory and be wrong whenever the source has moved on. A model that has the source in context will read it and be right.
Related terms
Parametric Knowledge
What the model knows from training, stored in its parameters. Frozen at training time.
Context
The relevant information the agent has access to right now. A measure of quality, not quantity.
Grounding
Tying the answer to something checkable, so it can be verified instead of trusted.
Hallucination
Output that is fluent and wrong. Two different problems hide under one word.
RAG
Fetch the relevant documents first, then let the model answer using them.
Want this working inside your team?
It starts with one conversation about your workflow.