What is context caching?
Saving reused prompt or context material so later model calls can be faster or cheaper.
03 · Context & memory · 17 terms
Context & memory covers what the system can see now and what it keeps for later. This section has 17 definitions. Open any term for a plain answer, the source behind it, and the day we checked it. Related links take you to the ideas that explain it or show where it sits in the wider field.
Saving reused prompt or context material so later model calls can be faster or cheaper.
Making working context smaller while keeping the facts and links a task needs.
The craft of choosing what a model sees, when it sees it, and in what form.
The instructions, facts, state, rights, and history given to one agent step.
The token limit shared by a request’s input and generated output.
Memory of past events or runs: what happened, when, and under what conditions.
Information kept across runs or sessions so an agent can use it later.
Information or state an agent keeps or finds again after its current reply.
Turning many short-lived records into a smaller set that lasts.
An attack that plants false or harmful facts in memory to sway later actions.
The selection of stored information for use in the agent’s current context.
A database or service where an agent keeps facts for later use.
Stored knowledge of how to perform a task, such as rules, routines, or reusable skills.
A method that finds outside facts and gives them to a model before it writes.
Memory of facts, ideas, and links without the full event where they were learned.
State kept for the current run or conversation, such as messages, tool results, plans, and checkpoints.
Goals, facts, and partial results kept close while an agent works.