Episodic Memory
An agent’s record of specific past events — what happened, when, and how it turned out.
Think of it like
A diary — not general knowledge, but “on Tuesday I tried X and it failed.”
Example
The agent remembers that last time it used a certain API it got rate-limited, so this time it slows down its calls.
How it actually works
Borrowed from psychology, episodic memory is about particular experiences rather than facts — a log of trajectories the agent can learn from. It lets an agent avoid repeating mistakes and reuse past solutions, but it also grows without bound, so you need retrieval to surface the relevant episode instead of dumping the whole history. Deciding what’s worth remembering is the hard part.
For product teams
Episodic memory is how an agent gets better with use instead of starting fresh every time.
For engineers
A store of past interaction traces, keyed for retrieval so relevant episodes can inform future decisions.
Related
- Working Memory — The fast, in-the-moment counterpart.
- Memory Retrieval — Pulling the right past episode back up.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome