VERA

Memory for
AI agents.

// Volitional Episodic Recall Architecture

VERA
02 / 14
The problem

AI has amnesia.

Every conversation starts from zero. Every agent forgets what happened yesterday — what mattered, what didn't, what to bring back.

Stateless intelligence is a commodity. Memory-aware intelligence is a moat. And nobody is modeling experiences over time — with emotional weight, decay, and intelligent retrieval.

VERA
03 / 14
The gap

RAG retrieves documents.
Vector DBs find chunks.
Nobody is modeling experiences.

Today's memory stacks are flat. Store a fact, search for it later. That's filing, not memory. Human memory decays, consolidates, and returns episodes with emotional context when the moment demands it.

VERA
04 / 14
What Vera is

An episodic memory layer between any model and any data source.

  • 13+ memory shape taxonomy — research-validated, modeled on human cognition
  • Dual-parameter power-law decay — important memories strengthen, trivial ones fade
  • Emotional gradient tracking — before, during, after — the full envelope
  • Volitional recall — the agent actively searches its own memory, not passive loading
  • Model-agnostic by design — Claude, GPT, Gemini, Llama, whatever comes next
VERA
05 / 14
Architecture

How Vera works

Any LLM Claude · GPT · Gemini · Llama
VERA memory layer
Episodic cognition · volitional recall · consolidation
Any data source Conversations · docs · signals
01 · Classify Three-axis scoring Emotional weight · context · deviation
02 · Encode Dual envelope Experience + standalone fact
03 · Consolidate Hippocampal triage Hot · warm · cold, nightly cycle
04 · Recall Volitional retrieval Agent chooses when to remember
01
VERA
§ 01 — The Thesis

Company,
not a feature.

VERA
07 / 14
Three reasons

Why episodic memory is a new category — not a LangChain plugin.

01

A new primitive

Episodic memory — with decay, emotion, and consolidation — requires building from scratch. If it were a feature, someone would have shipped it.

02

Platform, not plugin

Any model plugs in. Features depend on one platform's survival. Platforms survive because everyone depends on them.

03

Many markets

One architecture — personal AI, enterprise, healthcare, CRM. A feature solves one problem. A platform creates markets.

VERA
08 / 14
Competitive landscape

Why Big Tech can't just build this.

We reverse-engineered Anthropic's Claude Code memory — arguably the most sophisticated AI agent in production. Seven engineering layers, still flat markdown files. No emotional awareness. No episodic classification. No decay.

Capability Claude 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
Model dependencyClaude onlyAgnosticAgnostic
VERA
09 / 14
Not vaporware

What's actually working.

Live
Production bot — Claudette Running daily since Feb 2026. Real conversations, real memory.
Shipped
Three-tier classification Every memory scored on weight, context, deviation.
Shipped
Dual-classification (F3) Hard facts auto-extracted from emotional conversations.
Nightly
Consolidation lifecycle Hot/warm/cold triage with circuit breakers.
Shipped
Volitional recall Agent actively searches its own memory.
Live
Kill switches + compliance holds Production-grade safety across every subsystem.

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

VERA
10 / 14
First vertical

Healthcare alone could sustain a company.

  • The problem — CMS penalizes hospitals up to 3% of total Medicare reimbursement for excessive 30-day readmissions. Millions per hospital, annually.
  • The solution — Episodic pattern detection catches declining patient trajectories before readmission.
  • The moat — HIPAA-compliant by architecture. Sensitivity levels and compliance holds built in, not bolted on.
$15K+
Average readmission cost.
Vera's per-patient cost is a fraction of one prevented event.

Active conversations with healthcare providers under MNDA.

VERA
11 / 14
Platform reach

One architecture. Four markets. Same core.

Personal AI

Agents that know you over time

Retention and trust. The difference between a tool you use and one you rely on.

Enterprise

Institutional memory across agent teams

Knowledge preservation. Turnover-proof reasoning. Shared organizational context.

Healthcare

Episodic pattern detection for patients

Readmission prevention. CMS-aligned savings. HIPAA-compliant by architecture.

CRM · Sales

Relationship intelligence beyond logs

Revenue acceleration. Emotional context your CRM never captured.

VERA
12 / 14
Moat

What's defensible.

Provisional patent Filed March 25, 2026 — 6 claims, priority date locked. Utility filing within 12 months.
Working production system Daily use, real data, measured performance. Not a prototype.
Research validation 9 independent research agents confirmed the architecture. A-MAC · A-MEM · CLS · ACT-R mapped.
Competitive analysis 15+ competitors analyzed across 6 categories. Differentiation documented.
Academic paper Full episodic architecture documented. Section 3 first draft complete.
Model-agnostic architecture Connector-based, compliance-ready by design. Any LLM plugs in.
VERA
13 / 14
RM
Founder

Ryan Meadows

FOUNDER · VERA MEMORY INC.

Built the system before writing the paper. Reverse-engineered Claude Code's memory stack, mapped 15+ competitors, filed the patent, shipped the production bot. Built first. Theorized second.

VERA
14 / 14

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.

ryan@veramemory.ai · veramemory.ai · why.veramemory.ai
VERA
§ ASK
The ask

Raising a pre-seed round to scale the memory layer.

Round size
$1.5–2M
Pre-seed. SAFE or priced, open to structure.
Equity
~15–20%
Founder-friendly valuation. Patent + shipping product.
40%
Engineering — scale the memory layer
25%
Healthcare pilot — first paid deployment
20%
Research — utility patent + paper
15%
GTM — enterprise + personal AI design partners

Lead or follow welcome. Founder reachable at ryan@veramemory.ai.

Vera Memory Inc.

Memory for
AI agents.

// Volitional Episodic Recall Architecture

Ryan Meadows, Founder ↓ scroll
The problem

AI has amnesia.

Every conversation starts from zero. Every agent forgets what happened yesterday — what mattered, what didn't, what to bring back.

Stateless intelligence is a commodity. Memory-aware intelligence is a moat.

The gap

RAG retrieves documents. Vector DBs find chunks. Nobody is modeling experiences.

Today's memory stacks are flat. That's filing, not memory. Human memory decays, consolidates, and returns episodes with emotional context when the moment demands it.
What Vera is

An episodic memory layer between any model and any data source.

Architecture

How Vera works

Any LLMClaude · GPT · Gemini · Llama
VERA memory layerEpisodic cognition · volitional recall
Any data sourceConversations · docs · signals
01 Classify3-axis scoringWeight · context · deviation
02 EncodeDual envelopeExperience + fact
03 ConsolidateHippocampal triageHot · warm · cold
04 RecallVolitionalAgent chooses
§ 01 — The Thesis

Company,
not a feature.

Three reasons

Why episodic memory is a new category.

01

A new primitive

Episodic memory — with decay, emotion, and consolidation — requires building from scratch. If it were a feature, someone would have shipped it.

02

Platform, not plugin

Any model plugs in. Features depend on one platform's survival. Platforms survive because everyone depends on them.

03

Many markets

One architecture — personal AI, enterprise, healthcare, CRM. A feature solves one problem. A platform creates markets.

Competitive landscape

Why Big Tech can't just build this.

We reverse-engineered Anthropic's Claude Code memory. Seven engineering layers — still flat markdown files.

Memory types
Claude Code4 flat text categories
Mem0Flat facts + graph edges
VERA3-tier cognitive classification
Emotional awareness
Claude CodeNone
Mem0None
VERAFeeling signatures · gradients
Decay model
Claude CodeNone — persist until pruned
Mem0Simple recency
VERAPower-law w/ emotional protection
Consolidation
Claude CodeFile cleanup
Mem0Basic overwrite
VERAHippocampal triage lifecycle
Model dependency
Claude CodeClaude only
Mem0Agnostic
VERAAgnostic
Not vaporware

What's actually working.

Live
Production bot — ClaudetteRunning daily since Feb 2026.
Shipped
Three-tier classificationWeight · context · deviation scoring.
Shipped
Dual-classification (F3)Hard facts from emotional moments.
Nightly
Consolidation lifecycleHot/warm/cold triage with circuit breakers.
Shipped
Volitional recallAgent searches its own memory.
Live
Kill switches + compliance holdsProduction-grade safety.
Built first. Theorized second. The paper documents what works — not what we hope will work.
First vertical

Healthcare alone could sustain a company.

$15K+
Average readmission cost.
Vera's per-patient cost is a fraction of one prevented event.
Active conversations with healthcare providers under MNDA.
Platform reach

One architecture. Four markets. Same core.

Personal AI

Agents that know you over time

Retention and trust. A tool you use vs. one you rely on.

Enterprise

Institutional memory

Knowledge preservation. Turnover-proof reasoning across agent teams.

Healthcare

Episodic pattern detection

Readmission prevention. CMS-aligned savings.

CRM · Sales

Relationship intelligence

Emotional context your CRM never captured.

Moat

What's defensible.

Provisional patentFiled March 25, 2026 — 6 claims. Utility filing within 12 months.
Working production systemDaily use, real data. Not a prototype.
Research validation9 independent research agents confirmed the architecture.
Competitive analysis15+ competitors analyzed across 6 categories.
Academic paperFull episodic architecture documented.
Model-agnostic architectureAny LLM plugs in. Compliance-ready by design.
RM
Founder

Ryan Meadows

FOUNDER · VERA MEMORY INC.

Built the system before writing the paper. Reverse-engineered Claude Code's memory stack, mapped 15+ competitors, filed the patent, shipped the production bot. Built first. Theorized second.

The ask

Raising a pre-seed round.

Round
$1.5–2M
Pre-seed. SAFE or priced.
Equity
~15–20%
Patent + shipping product.
40%
Engineering
25%
Healthcare pilot
20%
Research + patent
15%
GTM

Lead or follow welcome. ryan@veramemory.ai

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.

ryan@veramemory.ai veramemory.ai why.veramemory.ai

Tweaks

Show ask slide

Hidden by default. When on, the ask appears as slide 15 — use it when the conversation goes there.