Vocabulary for building AI agents that plan, act, and recover from failure in production systems.
Information an agent can store and retrieve across steps or even across sessions. Typically categorized as in-context memory (inside the prompt), external memory (a database), and episodic memory (records of past interactions).
Deliberately designing what goes into the model's context — system prompt, retrieved docs, tool outputs, prior turns — rather than treating the prompt as an afterthought.
The maximum amount of text (measured in tokens) an LLM can process in a single call. Everything the agent knows about the current task must fit inside this window.
A fixed-length list of numbers (a vector) that encodes the meaning of a piece of text so that semantically similar texts end up with numerically similar vectors.
A pattern where relevant documents or data are fetched from an external store at query time and injected into the context window, so the model can answer using up-to-date or domain-specific knowledge it wasn't trained on.
The planned or enforced limit on how many tokens an agent is allowed to use across a task — balancing capability (more tokens = more thinking) against cost and latency.
A database optimized for storing and searching high-dimensional numerical vectors. In agentic systems it's used to find semantically similar documents quickly — the backbone of most RAG implementations.