Episodic Memory

What is episodic memory?

The term comes from psychology: besides remembering facts, humans remember "what I went through that day". In AI, episodic memory records the specific events an agent has lived — when, what it did, and how it turned out. Think of it as a personal log of experiences.

How is it different from semantic memory?

Semantic memory holds "what is"
"Paris is the capital of France" — factual knowledge, detached from time and place.
Episodic memory holds "what happened"
"Yesterday I queried the database with tool A and it timed out" — an experience with time and context attached.

What does it store?

The course and outcome of a task
Which task it took, which tool it called, whether it worked or errored out.
Key decisions and feedback
Why it chose that path, whether the user was happy — the lessons worth keeping on file.

Why it matters

With episodic memory, an agent stops being amnesiac. The next time a similar situation comes up, it can look back: last time this failed, so this time try another route; last time the user praised that style, so keep using it. Experience compounds, and the agent gets better with use.

Bottom line: episodic memory is the AI's "experience diary" — a record of what it did and how it went, so it learns from its own history.