Skip to main content

Engramme

Category: AI Infrastructure

Building Large Memory Models (LMMs) to endow humans with perfect and infinite memory by storing and proactively retrieving a person's entire digital life Engramme was founded in 2025. The company is led by Gabriel Kreiman. Based in San Francisco, CA, United States. Team size: 2-10. Total funding raised: $3M. Latest round: Pre-Seed. Key investors include Mayfield Fund, other undisclosed Silicon Valley investors.

Founded
2025
Headquarters
San Francisco, CA, United States
Team size
2-10
Total funding
$3M

Value proposition

Engramme is developing AI-powered memory infrastructure — Large Memory Models — that store and retrieve personal information like the human brain does, enabling proactive, searchless, associative recall of every conversation, document, and experience across a person's lifetime at petabyte scale.

Products and solutions

Large Memory Models (LMMs) — a new AI architecture beyond Transformers, Memory API (public beta) for embedding proactive memory into apps/services, Consumer iOS app (beta), Integrations with Gmail, Zoom, WhatsApp, Slack, Google Docs, wearables (Meta glasses)

Unique value

Neuroscience-grounded AI architecture (not Transformers) that mimics hippocampal memory encoding — proactive, associative, lifelong recall without search queries or prompts; hallucinations structurally impossible in memory layer

Target customer

Product teams and app developers (B2B via API); consumers (iOS beta app); enterprise teams needing organizational memory preservation

Industries served

AI Memory Infrastructure, Personal Productivity, Enterprise Knowledge Management, Healthcare (memory loss support)

Technology advantage

Proprietary Large Memory Model architecture built on 25+ years of peer-reviewed neuroscience research on how the brain encodes and retrieves memories; non-Transformer architecture with sparse attention, top-k memory routing, and document-wise RoPE; three core properties: lifelong petabyte-scale storage, proactive retrieval, associative recall

How they differentiate

Unlike competitors that patch LLMs with vector databases and embedding models (search-based memory), Engramme builds a fundamentally new AI architecture inspired by hippocampal neuroscience — proactive (not query-driven), associative (not keyword-match), and lifelong at petabyte scale. Claims hallucinations are structurally impossible in their memory layer.

Main competitors

Mem0, Rewind AI, Zep, LangMem, MemGPT

Key partnerships

Samsung, Superhuman, Dropbox, and Microsoft have expressed interest in embedding Engramme's memory API (per TestingCatalog), Mayfield Fund (lead pre-seed investor)

Major milestones

2025: Company founded by Gabriel Kreiman and Spandan Madan, Raised $3M pre-seed led by Mayfield Fund, 2026 (March): Emerged from stealth, Published Engramme Manifesto, 2026 (April): Seeking $100M raise at $1B valuation (per Bloomberg), 2026 (April): Memory API launched in public beta, Consumer iOS app in beta, Published "Memory is Not Search" position paper at ICLR 2026

Market positioning

Early-stage pioneer in the emerging "AI memory" vertical; positioning as the infrastructure layer for human-like memory in AI systems; seeking aggressive $100M raise at $1B valuation to establish category leadership

Geographic focus

United States (San Francisco / Boston); global market for AI memory infrastructure

About Gabriel Kreiman

Professor at Harvard Medical School (~20 years); Associate Director, MIT-Harvard Center for Brains, Minds and Machines; PhD in Biology, Caltech (2002, supervised by Christof Koch); Postdoc in AI with Tomaso Poggio at MIT; Author of 160+ research articles and 3 books

Official website: