Durable Memory
Agents that remember decisions and compound knowledge over time.
What it is
Instead of starting from zero in every thread, OpenClaw agents read and update persistent markdown state files (like `MEMORY.md` or `STATE.md`). They use automated memory compaction scripts to continuously archive lessons learned and flush stale context.
Why it matters
Agents without memory repeat the same mistakes. Durable memory allows your AI operators to compound their context over weeks and months, reducing token costs and ensuring that architectural decisions and standard operating procedures (SOPs) are strictly followed.