Skip to content
Home › Context Management & Reliability

Structured Note-Taking

Context Management & Reliability

Definition

A long-horizon technique where "the agent regularly writes notes persisted to memory outside of the context window. These notes get pulled back into the context window at later times" — external, agentic memory that survives context resets.

Key points

  • Structured note-taking (verbatim): "the agent regularly writes notes persisted to memory outside of the context window. These notes get pulled back into the context window at later times."
  • Provides "persistent memory with minimal overhead"; lets agents "track progress across complex tasks, maintaining critical context and dependencies" (e.g. a to-do list or NOTES.md).
  • Example cited: an agent maintaining state/notes across a long task and resuming from them after a context reset.
  • Distinct from Server-side Compaction (which summarizes in-context history) — note-taking persists information outside the window entirely.
  • The Memory Tool (memory_20250818) is the product-level implementation of this pattern (file-based /memories).

Why it matters for the exam

  • The concept behind scratchpads, NOTES.md/to-do files, and the Memory Tool. Expect long-running agent scenarios where progress and dependencies must survive a context reset.

Common gotchas

  • Note-taking is not compaction: it writes durable notes outside context rather than summarizing what's inside it — the two are complementary, not the same.
  • Notes only help if they're pulled back in later; the pattern requires the agent to re-read its own memory (cf. the memory tool's "view memory first" protocol).

See also

Sources

Referenced by