Membria.AI · Investor Deck

AI agents that learn.

Membria is the memory & experience layer for AI agents — one causal-memory core powering products for software teams, industrial engineering and audit.

02

Problem

Intelligence that resets every morning

1.

AI agents are stateless

Every session starts from zero. Yesterday's reasoning, decisions and failures are discarded the moment the window closes.

2.

Mistakes repeat — and they're expensive

Coding teams re-solve the same problems; in engineering, 28% of budget overruns are caused by rework.

3.

Handovers bleed context

30–50% of assumptions vanish at every phase handover; 18% of working time is spent searching for data that already exists.

4.

Judgment walks out the door

A senior engineer leaves — and fifteen years of decisions, lessons and "don't do this" leave with them.

Every AI tool today can answer questions. None of them remembers why decisions were made — or what happened next.

03

Why now

The experience graph is unclaimed

1.

Agents became the workforce

Claude Code, Codex and Cursor now do real daily work in code, engineering and finance — but they shipped without memory.

2.

MCP became the standard port

One protocol plugs a memory plane into every agent on the market — distribution without integration projects.

3.

Regulation demands traceability

EU AI Act and ISO 19650 require defensible decisions and immutable audit trails in safety-critical work — exactly what a causal memory produces by default. Europe is where this lands first.

4.

Incumbents own data, not experience

Nobody owns the decision → outcome graph. Whoever captures it first gets a moat that compounds with every user, every day.

04

Product

Membria Reasoning Graph Workforce

Institutional knowledge Memory pillars

Memory

Decisions — every choice captured with its reasoning
Consistency — behavior chains keep teams aligned
Skills — wins distilled into reusable know-how
NegativeKnowledge — failures become rules

Multi-agent harness Reasoning pillars

Reasoning

Mesh — many agents work in parallel
Coordination — agents hand off via A2A
Routing — each task to the right model
Quality gates — verification blocks bad output

Domain-agnostic foundation Graph pillars

Graph

GraphRAG — semantic search over a property graph
Multi-domain — one schema across every domain
Orchestration — work runs as a dependency graph
Adaptive — memory that strengthens with use

05

How it works

One loop, every stage

Not a memory bolted onto one step — every shipped outcome feeds the next decision.

01

Plan

Recalls decisions and outcomes from similar work — the agent starts from what already worked.

02

Execute

Injects validated skills and blocks known anti-patterns as the work is being done.

03

Review

Checks the result against past failures and standards — deterministic critics, auditable output.

04

Ship

Records the real outcome and links it back to the decision that caused it.

↺ Every outcome feeds the next decision. One MCP connection: npx @membria/cli init

06

The wedge

EPC first: nobody remembers

Description → P&ID in minutes

Plain words or an equipment CSV in; a laid-out diagram out.

Real CAD files, not pictures

DXF for AutoCAD, DEXPI for AVEVA/Hexagon, IFC4 for BIM.

Five-discipline critics

Process, Mechanical, HSE, I&C, Piping — deterministic, auditable, repeatable.

Every defect becomes memory

Finding → change request → negative knowledge, automatically.

Zero competitors with persistent memory

Autodesk, Bentley, AVEVA, Trimble, Procore — all stateless.

Regulation is the sales agent

EU AI Act + ISO 19650 make traceable decisions mandatory in safety-critical infrastructure.

Graph data model

Every P&ID, spec and decision — one connected graph

07

Data moat

The engineering knowledge graph no one else has

20,000+

engineering standards processed — API · ASME · IEC · ISO · ГОСТ · DNV · NFPA · AWS · ASTM

2.5M+

machine-extracted rules: ~1.2M prohibitions + ~1.3M best practices

10,000+

equipment types mapped to rules across 20+ disciplines

6

jurisdictions in one graph: US · EU · UK · Norway · CIS · International

ISO

~35,000

IEC

~10,900

ASTM

~9,200

ГОСТ (CIS)

~1,800

API

~1,760

ASME

~1,200

NFPA

~760

DNV

~700

AWS

~450

+ ACI · EEMUA · WRC · HI · FM · CGA · NORSOK · Eurocode — bars on a sqrt scale for readability

No cold start, ever. The same schema ships pre-loaded in code too — 96.5K production repos mined, ~570K Skills and AntiPatterns live. A generic memory startup begins every client at zero; we begin already knowing what fails.

08

Competition

Nobody remembers

Capability	Membria	Autodesk	Bentley	AVEVA	Procore	Symphony AI	Palantir Foundry	mem0 · Zep Letta
Persistent memory across sessions	✓	—	—	—	—	—	~	✓
Decision → outcome provenance	✓	—	—	—	—	—	~	—
Negative knowledge — what already failed	✓	—	—	—	—	—	—	—
Verification & quality gates	✓	~	~	~	—	—	~	—
Standards graph — ISO · API · ASME	✓	—	—	—	—	—	—	—
EPC domain depth	✓	✓	✓	✓	✓	~	~	—
Works with any agent via MCP	✓	—	—	—	—	—	—	✓
Multi-agent orchestration	✓	—	—	—	—	—	✓	—
Experience compounds across projects	✓	—	—	—	—	—	—	—
Serves small & mid-size business	✓	✓	~	—	✓	—	—	✓
Installed base & distribution	~	✓	✓	✓	✓	~	✓	~

✓ yes · ~ partial · — no · The real competition is inertia: Excel + "it works fine as is." Every project starts from a blank page — that is the memory problem, and it's solvable.

09

Traction & model

Sequencing, not spreading

CE subscriptions for solo developers · per-seat team plans · self-hosted enterprise license · EPC per-project pricing.

Now

Foundation

€500K revenue by end of 2026

1 bank + 1 government, live
EPC pilots in engineering
CE & EPC SaaS launching in parallel
100+ seats in production

Next 12 months

EPC revenue

€2M revenue in 2027

first EPC contracts
Glass audit beta
commons → 1–2M nodes
ISO 27001 / SOC 2 path

24 months

Memory-plane API

€6M revenue in 2028

open experience layer
any agent, any vendor
cross-industry commons
platform economics

10

Let's build the layer
agents can't work without.

Team & Ask

Let's Work Together

Philippe Khomenok

CEO · Co-Founder

Strategic vision, go-to-market and growth

Michael Aprossine

CTO · Co-Founder

Platform architecture, enterprise delivery

Mike Keer

Lead ML Architect

Memory & reasoning systems — fast, reliable, built to scale

Ask	Detail
Traction	1 bank + 1 government live enterprise · multiple EPC engineering pilots · 100+ seats deployed · CE & EPC SaaS launching in parallel
European Base	Belgium entity · relocating to Luxembourg · LuxProvide / MeluXina HPC partner · built for EU AI Act & ISO 19650 compliance
Raising	€500K pre-seed on a €5M pre-revenue valuation
Use of funds	EPC go-to-market · core team to 7 · 18 months runway to Series A
Contact	hi@actiq.ai · membria.ai · code.membria.ai · epc.membria.ai · glass.membria.ai (soon)