AI Agents · Const

Episodic Memory

Episodic memory captures past events and actions so an agent can remember how tasks were accomplished. It is commonly reused as few-shot examples that guide future task execution.

How it works

Episodic memory records past events or actions, especially the experiences behind successful task completion. Agents commonly reuse these records as few-shot examples, allowing future behavior to reflect how similar tasks were handled rather than only retaining isolated facts.

Key points

It focuses on retrieving what happened before, including prior actions taken by an agent.
Experiential knowledge belongs in episodic memory, in contrast to standalone factual knowledge.
Its practical value is preserving task-solving approaches that an agent can apply again.
Past action sequences can serve as demonstrations for improving later task execution.
As a long-term memory category, episodic information can persist beyond an individual conversation session.

Common mistakes

Treat episodic memory as experience examples, not merely a fact store; the documented pattern is to place prior sequences into few-shot prompts.

Example: Reuse an episodic example

import { ChatOpenAI } from "@langchain/openai";
import { HumanMessage } from "@langchain/core/messages";

const model = new ChatOpenAI({ model: "gpt-4o-mini" });
const examples = [
  { input: "Reset my password", approach: "Verify identity, then send a reset link." },
];

const prompt = `Use this past experience as a guide:\n${JSON.stringify(examples)}\n\nUser: Reset my password`;
const response = await model.invoke([new HumanMessage(prompt)]);
console.log(response.content);

A previous interaction is stored as a few-shot example and inserted into a later prompt so the agent can reuse the demonstrated approach.

When to use it

Support-task replay

An agent handles recurring support requests and reuses successful prior interaction sequences.

Few-shot task guidance

An agent selects previous task demonstrations to guide a new but similar workflow.

Cross-session learning

An application preserves experiences across conversations for later reuse.

Interview questions

QWhat is episodic memory, and why is it useful for agents?

Episodic memory is memory of previous events or actions. In an agent, it is useful for remembering how to accomplish a task, often by reusing past sequences as few-shot examples.

Follow-up: How does it differ from semantic memory? Semantic memory stores facts, while episodic memory stores experiences and task trajectories.

QHow is episodic memory commonly implemented?

A common implementation stores prior successful interaction sequences and includes selected examples in a later prompt. This gives the model demonstrations of an effective approach to imitate.

Follow-up: What should retrieval optimize for? It should select examples relevant to the current task rather than indiscriminately adding old interactions.

QHow does episodic memory differ from semantic memory?

The distinction is based on content: semantic memory captures facts, whereas episodic memory captures experiences. Episodic records therefore help an agent reproduce a method, not just recall information.

Follow-up: Can one interaction contain both types? Yes; its factual outcome can become semantic memory while the sequence and method remain episodic memory.

Semantic memory
Procedural memory
Long-term memory
Few-shot example prompting

Source: Memory overview. Checked 2026-10-15.