An LLM is amnesiac — each call starts blank. Memory is engineering: deciding what to keep, where, and what to re-inject. Critical for any long-running agent.
⚡Your agent forgets everything between sessions — re-asks what it already knew, and contradicts last week's answer.
LEVEL 1 — REMEMBER · the vocabulary
⚓ ANCHOR · IF YOU ALREADY KNOW
Agent memory is session state you'd build anyway: an append-only log of facts, a summarizer to keep it bounded, and a retrieval index to pull the relevant bits back. It's RAG pointed at the agent's own past.