Engramme Wants to Give You Infinite Memory — and That’s the Problem

A Harvard lab quietly launched a company called Engramme with a mission statement that should make every reader stop scrolling: “to endow humans with infinite memory.” You’ll recall every person you ever met. Every conversation. Every place you visited. Searchless. Promptless. You won’t have to ask — the memory will be there.

It’s a beautiful pitch. It’s also worth examining before we hand it our lives.

Who’s behind it

The founders aren’t lightweights. Gabriel Kreiman (CEO) is a Harvard neuroscience professor with roughly 160 publications and twenty years of work on how the brain encodes memory. Spandan Madan (CTO) holds a Harvard–MIT joint PhD and has shipped ML systems at Google DeepMind, Meta, Adobe, and Fujitsu. Two more founding researchers round out the team. Their published work spans Nature Human Behaviour, Cell Reports, NeurIPS, and ICLR, with papers on how the brain segments continuous experience into discrete memories, how sparse neural representations enable continual learning, and how machine-learning models can predict which moments humans will actually remember.

That’s not vaporware. That’s a serious lab spinning a serious product out the door.

What an engram actually is

The name is a tell. In neuroscience, an engram is the physical trace a memory leaves in the brain — the specific pattern of synaptic changes that lets you recall your grandmother’s face or the smell of a school hallway. Karl Lashley spent thirty years trying to find one and mostly failed; modern work using optogenetics has identified specific neurons whose activation can replay a memory in mice. Engramme is staking the company name on the idea that this trace can be modeled and externalized.

The technology underpinning a system like this is increasingly likely to involve a Large Memory Model (LMM) — a transformer-style AI architecture that, unlike a standard chatbot, carries a persistent memory bank it can read from, write to, and selectively forget. Whether Engramme’s specific approach uses LMMs, vector retrieval, or something purpose-built isn’t public yet, but the broader category is what makes “remember everything” newly plausible.

The thing nobody is saying out loud

Here’s where I want to slow down.

We already have systems that store memories. Writing. Photographs. Voice memos. Calendars. Search histories. Email. The entire architecture of your phone is a memory aid. Externalizing memory isn’t a new project — humans have been doing it for five thousand years, ever since the first cuneiform tablet recorded a beer transaction.

But every existing memory aid shares one feature: you have to deliberately use it. You decide a moment is worth photographing. You write the note. You search the email. The deliberate act of capture and recall keeps your own brain in the loop.

What “searchless, promptless recall” proposes is different. The capture is automatic. The retrieval is automatic. You never have to ask. Which sounds wonderful — and is exactly the part to think hard about.

The cost we’re not pricing in

Cognitive science has a term for what happens when we systematically delegate a mental task to a tool: cognitive offloading. Some of it is fine — you don’t need to memorize the multiplication tables your calculator can do faster. But the brain follows a use-it-or-lose-it rule. Spatial-navigation skills measurably decline in heavy GPS users. Phone numbers used to live in your head. Now you don’t know your spouse’s by heart.

There’s a related phenomenon called transactive memory — couples and teams that share knowledge develop divisions of cognitive labor. You remember the schedule, your partner remembers the names, neither of you has to do both. Healthy. But when one half of the partnership is a piece of software with infinite capacity, the human side can stop carrying its share of the load entirely.

The risk with infinite memory isn’t dystopian. It’s quieter. It’s that the act of trying to recall — the small struggle of “what was that name?” or “what did she say at dinner?” — turns out to be the act that builds and maintains the memory in the first place. Take that struggle away and the underlying capacity withers. Not all at once. Over years. The way muscle atrophies in a cast.

This is not a hypothetical. The research literature on hyperthymesia — the rare condition of “highly superior autobiographical memory” — is full of people who can recall every day of their lives and frequently report that the experience is exhausting, intrusive, and emotionally flattening. Forgetting is a feature, not a bug.

Two paths

There’s a way to use AI that enhances and augments what makes us human. Tools that remind us at the right moment, surface a name we’re searching for, suggest the connection we almost made. The brain still does the work; the AI extends the reach.

There’s a second path where AI gradually replaces those things. We stop attempting to recall. We stop noticing what’s worth remembering. We stop the small, daily, neurologically vital act of trying. The AI becomes the memory and the human becomes the consumer of its output.

Engramme might land on either side of that line. The team is good enough that the product will probably work. Whether it’s good for the user is a separate question, and one the marketing copy isn’t asking.

The specific next step

If you’re going to try a “remember everything” service — Engramme or any of the dozen quantified-self tools coming behind it — apply this test before you commit. Does the tool still require you to do the act of remembering? Or does it remove that act entirely?

A note app that lets you search your own writing keeps you in the loop. A camera roll you periodically scroll through keeps you in the loop. A passive recorder that captures everything and surfaces it without prompting does not.

Pick the tools that make you a more capable rememberer. Be skeptical of the ones that offer to do the remembering for you.