Memory Guide

Safe long-term memory handling for AI agents requires rule-driven guardrails that dictate exactly when agents can read or edit memory across sessions. This ensures traceability and governance by preventing off-pipeline writes.

The memory pyramid structures AI agent memory into T0, T2, and T3 flows with heartbeat curation and soul.md integration. It provides a tiered governance framework to securely manage long-term contextual data.

Manual memory interventions are appropriate when you need to persist important facts deterministically. You perform these direct writes by using the save_memory function with the appropriate category.

You can bypass heartbeat filtering only by following a strict safe override protocol. This ensures that governance, security, and traceability are maintained when deviating from standard memory curation.

Preventing off-pipeline memory writes involves defining specific frontmatter fields, roles, and usage guidelines. This enforces deterministic writes and stops unauthorized memory modifications outside the pipeline.

Implementing memory governance requires specifying frontmatter fields and roles to define usage guidelines. This structure enforces deterministic saves and secures interactions with the memory pyramid.