iCog by CognitivX

Confidential investor material · May 2026

Persistent memory. Continuous cognition.

Not smart. Just knows you.

iCog gives people a single memory graph across the AI tools they already use, then runs cognition on that graph to produce recall, patterns, and personal context no single assistant can hold alone.

Liveicog.app · API · MCP

$500KPre-seed SAFE planned

$8MPost-money cap target

Problem

Every assistant starts from zero.

ChatGPT does not know what Claude learned yesterday. Cursor does not know what a user told Copilot. Every new agent asks the user to reconstruct context again.

Context tax Users re-explain their work, preferences, people, and projects in every tool.

Insight loss No single assistant sees the cross-tool pattern of a person's AI life.

Vendor lock-in Built-in memory works only inside one lab's product boundary.

Solution

A portable cognition layer for AI tools.

User

iCog memory graph
typed memories · vector recall · identity anchors

ChatGPT · Claude · Cursor · IDE agents · MCP clients

Remember

Store user-owned memories from conversations, tools, agents, and API clients.

Recall

Return the right context at the right depth: foundational, standard, deep, identity.

Reflect

Run nightly consolidation over the graph — clustering, community detection, and insight synthesis — turning scattered conversation into durable patterns.

Why now

The missing layer became technically and commercially urgent.

01

MCP matured

Cross-assistant context can now be distributed without one-off integrations for every tool.

02

Memory moved from novelty to need

Frontier assistants proved users value memory, but each implementation remains vendor-bound.

03

Hardware capture lost trust

Always-on recording pushed privacy-sensitive users toward software-first, user-owned memory.

04

Consumer AI monetizes fast

Users already understand the $20 and $200 AI subscription bands.

Product

One memory substrate, four surfaces.

icog.app Consumer chat and PWA

console.cognitivx.io Developer and billing console

api.cognitivx.io OpenAPI memory and recall API

mcp.cognitivx.io MCP server for AI tools and agents

Production-live Stripe-billed tiers OpenRouter model routing PostgreSQL + pgvector

Demo

Not a prototype. A live system you can use today.

The live iCog consumer product at cognitivx.io / icog.app — cognitivx.io · icog.app — the live consumer surface: remember, connect, reflect, recall.

iCog public OpenAPI documentation — api.cognitivx.io/docs — the public memory & recall API (OpenAPI 3.1), in production.

Live front-end Documented public API MCP server for any AI client

Moat

The database is not the moat. The cognition loop is.

Defensible asset The cognition loop

Raw data commoditizes; vector search is a library import. The moat is the engineered loop that runs on a cross-vendor graph — activation-ranked recall feeding nightly consolidation feeding richer recall. It is shipped, running code a competitor has to rebuild, and accumulated personalization compounds into switching cost.

Activation-ranked recall Reciprocal-rank fusion over vector + keyword, weighted by Ebbinghaus memory decay and PageRank centrality — not raw cosine similarity.

Dream consolidation 35+ nightly jobs — clustering, community detection, learning-to-rank — synthesize durable insight. Structure first; the model only renders language.

Cross-vendor by design No single-vendor memory can follow a user across tools without cannibalizing its own lock-in. That asymmetry is structural, not temporary.

Market

A $467B AI software market — and a personal memory layer no incumbent owns.

$174B Global AI software market, 2025 · ABI Research

$467B Projected 2030 at a 25% CAGR · ABI Research

$7.8B → $53B AI agents market, 2025–30 at ~46% CAGR · MarketsandMarkets

~$32M ARR iCog base case by Y5 · ~48K paid users, under 1% of paying AI users

TAM: global AI software market — ABI Research (2025). Adjacent layer: AI agents market — MarketsandMarkets (Apr 2025), triangulated by Grand View and Precedence. The personal, cross-tool AI-memory layer is a green-field category with no incumbent third-party market size — iCog is defining it. SOM is bottom-up from the iCog base-case financial model (~48,345 paid users, ~$32M ARR by Y5).

Business model

A single credit meter across subscriptions, PAYG, API, and partners.

TierPriceCreditsMemory

Free$0100/mo1K memories

Awakened$20/mo500K/mo1M memories

Conscious$200/mo15M/moUnlimited

PAYG$0.101K creditsUnbounded

Blended ARPU model: $52/mo Gross margin model: 81-83%

Usage-based memory pricing is proven at scale: Mem0 grew to 186M memory API calls and raised $24M in 2025 on operation-metered pricing (Sacra, TechCrunch). iCog’s PAYG + API meter captures the same agent-driven usage — revenue scales with operations, not seats, which is the upside lever beyond the consumer tiers.

Competition

Infrastructure remembers for apps. iCog remembers for people — and their agents.

Traditional apps needed a database. Modern agents need cognitive memory.

CategoryExamplesGapiCog position

Memory infraMem0, Letta, ZepDeveloper-firstPeople + their agents

Model memoryOpenAI, GoogleSingle-vendorCross-vendor

Second brainNotion, ObsidianUser-input heavyLearns from AI use

Capture hardwareRewind, Bee, FriendPrivacy burdenSoftware-first

Why won’t the labs just do this? Single-vendor labs are structurally disincentivized from portability — cross-tool memory cannibalizes the lock-in their valuations depend on. The neutral, user-owned layer is the one place they won’t follow.

Status

Production is live. The round funds distribution and proof.

Built

Consumer app, API, MCP server, billing console, docs, architecture stack, and business-plan model.

Verified

Live Stripe pricing, Vercel surfaces, Cloudflare DNS, OpenRouter routing, and generated API docs.

Next proof

Convert pre-launch product into repeatable acquisition, retention, and paid-user cohorts.

Month 3: channel-quality gate Month 6: NRR >= 100% Month 12: $1M ARR path

Team

One founder shipped the entire cognitive stack.

Founder Parsa Barati

Founder & sole developer, CognitivX. 22, building from the UK; previously built MoltJobs and Lexaplus. Solo-designed and deployed the full iCog system — a 300+ module cognitive engine, public API, MCP server, billing, and a live consumer app — over seven months of cognitive-architecture research since October 2025.

Shipped solo 300+ module cognition engine, 50+ API routers, MCP server, Stripe billing, and four live product surfaces.

Depth, not glue Memory model, blended recall, and a 35-job nightly consolidation engine — built from the cognitive-architecture literature, not assembled from templates.

What the round adds First hires in growth and applied cognition — turning a proven engine into adopted product.

Sole founder & developer 7 months R&D · since Oct 2025 Prior: MoltJobs · Lexaplus

Financials

Base case: efficient pre-seed path to $30M+ ARR.

YearRevenueARR EOYPaid usersGM

Y1$341.5K$589.8K94181.5%

Y2$1.49M$2.12M3,35282.2%

Y3$4.53M$6.20M9,61682.6%

Y4$10.81M$13.98M21,29182.8%

Y5$24.64M$32.13M48,34583.0%

Base scenario from the iCog / bp-site financial model; blended ARPU ~$52/mo. Generated May 2026.

Vision

Today we remember for you. Next we predict what you’ll need.

Today — shipped

The cognition loop: activation-ranked recall and nightly consolidation over a cross-tool memory graph. Running in production.

Next — Foresight

A personal world model that simulates outcomes forward. The delta between choices is the product — not absolute predictions.

Horizon — the cognitive layer

Predict the user’s latent state and needs across every tool and agent: the runtime that sits above all foundation models.

Grounded in 40 years of cognitive-architecture research — ACT-R, SOAR, Global Workspace, predictive coding, and the modern world-model literature. We model the user’s state, not the next token.

Shipped: v11–v14 cognition engine Designed: v15 Foresight world model Memory is the wedge · cognition is the company

Ask

$500K pre-seed SAFE to turn a live product into a funded category wedge.

Round $500K SAFE

$8M post-money cap · 20% discount · MFN · pro-rata

Use of funds 60 / 25 / 10 / 5

Growth · R&D · operations and governance · reserve

Entity path CognitivX Ltd.

UK product entity planned at pre-seed close, with IP assignment from Black Megaurion Ltd.

Parsa Barati · Founder, CognitivX parsa@cognitivx.io