Memory — how agents remember
Memory lets an agent carry what it learns from one run to the next — your preferences, ongoing context, and decisions — so it gets better the more you use it.
Last updated July 14, 2026
Memory is what lets an agent carry context from one run to the next. Instead of starting fresh every time, it can remember your preferences, ongoing details, and decisions you've made together — so it gets more useful the more you work with it.
What agents remember
Memory captures the kinds of things a good assistant would hold onto:
- How you like things done — your preferred tone, formats, and defaults.
- Ongoing context — the project you're in the middle of, who the key people are.
- Decisions and corrections — feedback you've given that should stick going forward.
How it builds up
As an agent works, it notes what's worth remembering and carries it forward. When you correct it or state a preference, that becomes part of what it knows for next time. Over weeks, this turns a capable agent into one that fits the way you actually operate.
Memory vs. Knowledge
They're easy to confuse. Knowledge is reference material you curate — stable facts you want the agent to answer from. Memory is what the agent learns as it works with you over time. Use knowledge for your source of truth, and let memory handle the accumulating context of your working relationship.
Managing what's remembered
Check what an agent has picked up so you know what's shaping its behavior.
If something it remembered is out of date or wrong, tell it — the correction replaces the old assumption.