Why AI Agents Need Memory - Interactive 3D

Memory console

Persistent memory Context window (working slots): 6

0Lessons retained across tasks

The reset problem

Most agents today are goldfish. A task starts, the model gets a context window full of instructions, it acts, the task ends — and everything evaporates. Run the simulator with Persistent memory off: each task deposits lessons (the amber beads), and each task-end wipes them. The retained counter never grows. The agent re-discovers the same failures forever, and every run costs the same tokens to re-learn.

Flip persistence on and the beads survive into the episodic ring. By task three, the agent starts from accumulated experience — that difference, not smarter reasoning, is what turns a tool-caller into something you can leave alone for a week.

The four memory types (borrowed from cognitive science)

Type	What it holds	Typical implementation	Lifespan
Working	The current task: instructions, recent tool outputs	The context window itself (8k–1M tokens)	One session
Episodic	What happened: past runs, failures, outcomes	Logs + vector store, retrieved by similarity	Weeks–forever
Semantic	Facts about the world and the user ("deploys happen Fridays")	Structured store / knowledge graph / notes file	Long-lived, edited
Procedural	How to do things: learned skills, playbooks, prompts that worked	Saved routines, fine-tunes, skill libraries	Long-lived, versioned

Worked example: an agent asked to “fix the failing build” should consult episodic memory (“last Tuesday the same error came from a stale lockfile”), semantic memory (“this repo uses pnpm, not npm”), and procedural memory (“the lockfile-repair playbook”) — and only load the relevant slices into its limited working memory. Try shrinking the context-window slider: with fewer working slots, retrieval quality matters more, not less.

Memory illustration capacity and durability boundaries

Your browser does not support video.

Read the explanation

The memory illustration represents working context as a ring of cubes. The native capacity slider ranges from two through ten, starting at six. Bars span one hundred fifty and two hundred fifty at twenty five pixels per drawn slot, comparing default six and maximum ten. Rebuilding the ring changes cube count, not a measured token limit or an actual model context window. Working, episodic, semantic and procedural shapes are illustrative categories. No language model, vector search, memory database or validated procedural retrieval runs inside this visualization. A changing capacity control demonstrates authored geometry only when its real renderer and listeners are available. The task animation chooses two plus the floor of a random value times two, producing two or three lesson beads with equal probability under its assumed uniform random generator. Bars show one hundred twenty and one hundred eighty pixels at sixty pixels per bead. The expected count is two point five per completed task, but an individual task can show either outcome. Beads have positions and appearance rather than semantic lesson content. A pulse callback appends them after an animation, and the task end later reads the bead array length. This is illustrative bookkeeping, not model learning, retrieval quality or measured sample efficiency. Overlapping runs or event timing require their own actual interaction evidence. The persistence switch determines whether a task-end callback retains the in-memory bead array or removes it. It never writes those beads to browser storage or a durable backend. Reloading initializes the array and retained counter to zero, even with the checkbox checked. Bars contrast an illustrative three beads held in memory and zero after reload at sixty pixels per count. The source constructs a Three renderer at the start of the application script before binding controls. Missing Three therefore leaves native slider movement disconnected from its caption and leaves the retained counter zero. Original blank positive canvas buffers and layout can be preserved, but that does not certify a functioning memory system, exhaustive interactivity or public playback.