"Is this a feature or a company?"

Company,
Not a Feature.

Vera is a new layer of infrastructure — episodic memory that sits between any AI model and any data source. Features live inside products. Platforms live underneath them.

The Thesis

Three Reasons This Is a Company

1

Episodic Memory Is a New Primitive

RAG retrieves documents. Vector databases find similar chunks. Nobody is modeling experiences over time — with emotional weight, decay, consolidation, and intelligent retrieval based on what happened and why it mattered.

This isn't something you bolt onto LangChain. It's a new category of AI infrastructure that requires building from the ground up.

  • 13+ memory shape taxonomy (research-validated)
  • Dual-parameter power-law decay modeled on human cognition
  • Emotional gradient tracking (before / during / after)
  • Consolidation cycles that strengthen important memories and let trivial ones fade
  • Dual-classification — emotional context and hard facts stored separately as the experience envelope AND the standalone searchable fact

If it were a feature, someone would've shipped it already. The reason nobody has is because it requires building from scratch — and we did.

2

Model-Agnostic = Platform, Not Plugin

Vera wasn't built for Claude or GPT. It's a memory layer that any model plugs into — Claude, GPT, Gemini, Llama, open-source, whatever comes next.

  • Features depend on one platform's survival
  • Platforms survive because everyone depends on them

Vera is the transmission, not the engine. Engines get replaced. Transmissions get integrated.

3

Multiple Markets Prove Platform Value

The same core architecture serves:

MarketUse CaseValue
Personal AIAgents that truly know you over timeRetention & trust
EnterpriseInstitutional memory across agent teamsKnowledge preservation
HealthcareEpisodic pattern detection for patientsReadmission prevention
CRM / SalesRelationship intelligence beyond logsRevenue acceleration

A feature solves one problem. A platform creates markets.

Competitive Landscape

Why Can't Big Tech Just Build This?

They can build "memory." They can't build this memory. We reverse-engineered Anthropic's Claude Code memory system — arguably the most sophisticated AI agent in production.

CapabilityClaude Code (Anthropic)Mem0 ($24M raised)Vera
Memory types4 flat text categoriesFlat facts + graph edges3-tier cognitive classification
ClassificationNone — stored identicallyNoneThree-axis scoring
Emotional awarenessNoneNoneFeeling signatures, gradients, shapes
Decay modelNone — persist until prunedSimple recencyPower-law with emotional protection
ConsolidationFile cleanup (dedup, prune)Basic overwriteHippocampal triage lifecycle
Fact extractionNoneNoneAuto-extracts from emotional moments
Model dependencyClaude onlyModel-agnosticModel-agnostic

Anthropic invested seven engineering layers in memory and still arrived at flat markdown files with a 200-line cap. No emotional awareness. No episodic classification. No decay. If they couldn't solve it with unlimited model access, memory infrastructure isn't a side feature. It's a company.

Claude Code built better RAM — managing what's loaded right now. Vera builds long-term memory — knowing what matters, strengthening important experiences, letting trivial ones fade.

Not Vaporware

What's Actually Working

This isn't a pitch deck with mock-ups. The system runs daily.

Live
Production bot (Claudette) Running since Feb 2026 — processing real conversations daily
Shipped
Three-tier classification Every memory scored on emotional weight, context dependency, and deviation
Shipped
Dual-classification (F3) Hard facts auto-extracted from emotional conversations
Shipped
Retrieval tracking (F1) Tracks active memory use, prevents false demotions
Nightly
Consolidation lifecycle Hot/warm/cold triage with circuit breakers and emotional protection
Shipped
Fact category taxonomy (F2) 7 governance categories — medical, financial, legal, relational, and more
Shipped
Volitional recall Agent actively searches its own memory, not just passive loading
Live
Kill switches On every subsystem — production-grade safety

Built first. Theorized second. The paper documents what works, not what we hope will work.

Moat

What's Defensible

Provisional Patent Filed March 25, 2026 — 6 claims, priority date locked. Utility filing within 12 months.
Working Production System Not vaporware. Daily use, real data, measured performance.
Research Validation Architecture validated against A-MAC, A-MEM, CLS theory, ACT-R. 17 prior art patents mapped.
Competitive Analysis Reverse-engineered Claude Code, Mem0, Zep, Letta, Cognee. Documented differentiation.
Academic Paper Section 3 first draft complete — full episodic architecture documented.
Model-Agnostic Architecture Connector-based, compliance-ready by design. Any LLM plugs in.
First Vertical

The Healthcare Proof Point

Healthcare alone could sustain a company.

If the healthcare vertical alone is a viable business, that tells you what the full platform is worth.

The Bottom Line

Features get absorbed.
Companies define categories.

Vera isn't adding memory to someone else's product.

Vera is building the memory layer the entire AI ecosystem needs.