Engram
by Engram
A learned memory layer that makes AI actually know your organization — at up to 100x fewer tokens.
Engram is a learned memory layer for enterprise AI: it trains a compact, reusable memory of a company's documents, code and communications in advance, so models answer with real institutional knowledge instead of re-reading context on every query. It is aimed at enterprises and AI product teams whose agents need deep organizational context without frontier-model token bills.
At a Glance
- Category
- AI Models & APIs
- Pricing
- Contact for pricing
- Target Market
- CTOs, CIOs, Heads of AI, Enterprise Developers, Data Scientists, AI Product Teams
- Founded
- 2026
- Headquarters
- San Francisco, USA
- Customers
- Early partners and customers include Microsoft, Notion and Harvey
Key Features
- ✓Cartridges (pre-trained memory)
Compresses a large corpus into a compact, reusable memory object the model loads at query time instead of re-ingesting raw documents.
- ✓Active Reading
Trains the model to study source material deeply ahead of time rather than merely storing and retrieving it.
- ✓Sparse memory fine-tuning
Lets the system absorb new organizational knowledge continuously while minimizing catastrophic forgetting.
- ✓Continuous memory refresh
The learned memory updates as company information changes — currently on a daily cadence, targeting hourly and finer.
- ✓Agent API for shared workspaces
An API designed for agents that operate over large, shared organizational knowledge bases.
Capabilities
Use Cases
- •Cost-efficient enterprise agents
Run agents over a company's full corpus while consuming a fraction of the tokens a frontier model would need for the same context.
- •Institutional-knowledge assistants
Build assistants that retain and reason over internal documents, code and communications rather than re-retrieving them each session.
- •Domain knowledge internalization
Used by Harvey to internalize legal knowledge and precedent, and by Notion to power custom agents over customer workspaces.
Ideal For
Best For
- ✓Cutting token spend on agents that need large amounts of organizational context
- ✓Giving enterprise AI assistants durable institutional knowledge instead of per-query retrieval
- ✓AI product teams whose agents operate over large shared knowledge workspaces
- ✓Replacing or augmenting brittle RAG pipelines with a pre-trained memory
Market & Ratings
Early partners and customers include Microsoft, Notion and Harvey
Market Analysis
Pros
- ✓Directly attacks the two biggest enterprise-AI pain points: context quality and token cost
- ✓Exceptional research pedigree — the team authored the Cartridges method
- ✓Microsoft, Notion and Harvey as named early partners is unusual traction for a company this young
Cons
- ✗Founded in 2026 and only months out of stealth — minimal production track record
- ✗The 100x-fewer-tokens figure is a vendor claim, not an independently benchmarked result
- ✗No public pricing, self-serve access or published SDK yet
Pricing
Enterprise
Contact for pricing
- ✓Custom trained memory per organization
- ✓Agent API
- ✓Continuous memory refresh
Pricing is not public; Engram works with enterprises directly.
Stay Ahead of the Curve
Weekly enterprise AI insights for technology leaders. No spam, no vendor pitches—unsubscribe anytime.
SubscribeRelated Products
TwelveLabs
Video intelligence API that makes every hour of enterprise footage searchable, analyzable and agent-ready.
Mistral OCR 4
Structure-aware document AI that returns bounding boxes, typed blocks, and per-word confidence scores.
Hugging Face
The AI community building the future.
Amazon Bedrock
The platform for building generative AI applications and agents at production scale.