← Back to Glossary
Agent Memory (Short-Term)
Phase 04 · Agentscore-concept

Working memory held in the LLM context window during an active session (scratchpad and recent tool outputs).

Why it exists

Allows the agent to maintain continuity across reasoning turns and tool invocations within one run.

How it works

Appends tool calls, observations, and thoughts to the prompt messages array passed to the API.

Where you'll use this