memorylake vs innernet

memorylake built a lake. innernet is building a life.

That isn't a jab — it's their metaphor and it's an accurate one. memorylake is enormous, multimodal, enterprise-grade memory infrastructure with bundled corpora and serious scale behind it. If your problem is volume of material, they are formidable and we are not competing for that job.

But there is one design choice at the centre of their product that we think is exactly wrong for personal memory, and it's worth being direct about: they resolve contradictions automatically. We keep them.

Persistent, multimodal memory infrastructure that works across models and agents — "the memory lake for every AI." Three surfaces: Memory Passport (a personal encrypted memory container), Workbrain (team productivity intelligence), and a Router for distributed systems. Ingestion covers conversations, documents, spreadsheets, audio, video and images. Six fixed memory types — background, fact, event, dialogue, reflection, skill. Git-like versioning with branching and time travel. Conflict detection with auto-resolution and audit trails. Tens of millions of built-in open datasets: academic papers, SEC filings, clinical trials, patents.

The scale claims are real data-platform claims and they come from a company with a data platform underneath. Believe them.

A personal memory layer that any AI tool can read. npx innernet connects the tools already on your machine to one memory. netti — one per person — files what accumulates into Context Maps: dimensions that read like documents, nodes for the people and decisions that recur, commits and branches. Every fact carries how far it may travel.

The target isn't scale of material. It's depth of understanding about one person.

This is the real argument and everything else is detail. memorylake detects conflicting memories and resolves them automatically. For an enterprise knowledge base that is unambiguously correct — two versions of a policy is a bug. For a person it is the opposite. You believed something in March and stopped believing it in July; a system that reconciles those into one current fact has quietly erased the most informative thing about you. innernet holds both, versioned, and treats the gap as signal.

A fixed schema — background, fact, event, dialogue, reflection, skill — makes memory tractable at ten-trillion-record scale. It also decides in advance what shape your life is. innernet lets netti design the dimensions a project actually needs and add more as the work reveals them. Our maps end up with dimensions nobody would have thought to put in a schema, because they came out of the work.

Bundled SEC filings and patent libraries are a fantastic asset — for research. They say nothing about you. The centre of gravity is unmistakable: memorylake's product gets better as the corpus grows. innernet's gets better as it reads more of your reasoning. Those roads don't converge.

• ·Multimodal ingestion. Spreadsheets, audio, video, scanned documents. We're text-shaped and that's a true gap.
• ·Scale. Their retrieval-at-volume claims come from a real data platform. We're not going to out-engineer that and we don't need to.
• ·Bundled corpora. Papers, filings, patents and clinical trials searchable alongside your own memory is something nobody else in this category offers.
• ·The enterprise surface. Workbrain, team RBAC and audit trails are a complete answer to a question we haven't tried to answer.
• ·They publish. Public benchmark numbers and comparison pages. We're catching up on that and they got there first.

• ·We keep what you changed your mind about. Branch it, park it, roll it back. Their versioning traces records for audit; ours forks a direction you work inside.
• ·Structure that grows into the shape of your work rather than six boxes chosen before you arrived.
• ·Disclosure on every fact. Encryption protects your memory from outsiders. Travel classes decide what the model is allowed to know — a different promise, and the one that matters when you connect an AI.
• ·netti is one per person. Presence and tone are part of the product when the subject is your own life, not a filing cabinet.

Use memorylake if you have a large body of material — documents, media, institutional knowledge — and you need an AI-accessible store over it, at scale, with enterprise controls.

Use innernet if the thing you want remembered is your own reasoning, across the AI tools you already use, including the parts of it that changed.

"They do git-like memory with branching too. Isn't that your whole thing?"

Same word, opposite intent. They version records so an auditor can trace them, then auto-resolve conflicts so the current state stays clean. We branch so a person can hold two incompatible directions at once and decide later — and we never auto-resolve, because a contradiction in your own memory is usually you changing, not an error to clean up. Theirs is provenance. Ours is thinking.

Accurate as of 3 September 2026, based on memorylake's public site. Their scale and benchmark figures are as claimed by them and we haven't independently verified them. If we've described their product unfairly, tell us and we'll correct it.