---
title: "What is episodic memory in AI agents?"
description: "Episodic memory lets an AI agent store specific dated events and recall what happened, with the time, context, and source of each event."
canonical: https://past.dev/glossary/episodic-memory
last-updated: 2026-09-02
---
# Episodic memory

Source: https://past.dev/glossary/episodic-memory

Episodic memory stores specific events with their time and context, so an agent can recall what happened rather than only what is generally true. Each entry records one occurrence: a call, a decision, an outage, with its date, participants, and outcome. AI agents use episodic memory to answer questions such as what a customer reported on March 12, with the source that proves it.

## What an episodic record contains

An episodic record describes one event. The term comes from cognitive psychology, where episodic memory names a person's memory of experienced events. In an agent memory system, the record keeps four things together:

- **What happened**: the content of the event, kept close to the original wording of the source.
- **When it happened**: the [event time](/glossary/event-time), taken from the source record's own timestamp rather than from when the system stored it.
- **Who was involved**: participants linked across records by [entity resolution](/glossary/entity-resolution), so 'Acme', 'Acme Corp', and 'the customer' resolve to one entity.
- **Where it came from**: the transcript, email, or ticket that records the event, so the memory can be verified.

## Why it matters for AI agents

General facts answer what is true. Episodic memory answers what happened, and many agent tasks turn on that. Preparing for a renewal call, explaining a decision, or checking what was promised all require specific events with dates. A rolled-up customer profile cannot answer them, because summarization removes the event level. Events also need order: [event ordering](/glossary/event-ordering) places them by event time rather than arrival order, so the sequence matches what happened.

Long-horizon benchmark suites such as [LongMemEval](/benchmarks/longmemeval) test episodic recall directly: they ask questions about specific events from earlier sessions and score whether the system retrieves the right one. A memory API such as [past.dev](/docs/memory-api/quickstart) stores each record with the source's own timestamp, so recall returns the dated events behind an answer along with their sources.

## Episodic memory in practice

A support agent picks up a ticket from Acme Corp on September 1. Acme reports intermittent API errors. Episodic memory holds the outage call from March 12, in which Acme's engineers reported the same error codes and an engineering lead promised a fix in the next release. The agent recalls that call, cites its date, and responds with the history: the problem recurred, and a commitment already exists.

Without the episodic record, the agent knows at most the general fact that Acme has had incidents. With it, the agent can quote the March 12 call, name what was promised, and check whether the promise was kept.

## Related concepts

- **[Semantic memory](/glossary/semantic-memory)**: general facts distilled from events, such as which plan Acme is on.
- **[Procedural memory](/glossary/procedural-memory)**: how tasks are performed, such as the refund procedure.
- **[Working memory](/glossary/working-memory)**: what the agent holds in the context window for the current ticket.
- **[Long-term memory](/glossary/long-term-memory)**: the umbrella term for everything kept outside the context window.
- **[Event time](/glossary/event-time)**: the timestamp an episodic record is indexed under.

## Frequently asked questions

### What is episodic memory in an AI agent?

Episodic memory is the part of an agent's memory that stores specific events with their time and context. It lets the agent recall what happened on a given date rather than only general facts.

### How is episodic memory different from semantic memory?

Episodic memory stores individual events, such as a call on March 12. Semantic memory stores general facts distilled from events, such as the plan a customer is on.

### Do AI agents need episodic memory?

Agents that work with people or cases over time need it. Questions about what was said, promised, or decided can only be answered from stored events with dates and sources.

## Related

- [Semantic memory](https://past.dev/glossary/semantic-memory)
- [Working memory](https://past.dev/glossary/working-memory)
- [Event time](https://past.dev/glossary/event-time)
- [Event time vs ingestion time](https://past.dev/guides/event-time-vs-ingestion-time)
- [Memory API quickstart](https://past.dev/docs/memory-api/quickstart)