Architecture

Not bots.
An organization.

42x isn't a chatbot or a workflow tool. It's a full management hierarchy that thinks in BDI cycles, governs itself with autonomy tiers, and learns from every action.

Hierarchy

A full management team.
Always running.

The CEO agent sets strategy every 4 hours. Venture Managers coordinate every 2 hours. Functional agents execute every hour.

Each agent has its domain, its tools, and its autonomy level. They communicate through tasks, directives, and shared context — not chat.

Agent Hierarchy
CEOStrategy & allocationevery 4h
Venture ManagerExecution & coordinationevery 2h
MarketingContent, SEO, campaignsevery 1h
SalesChannels, pricing, outreachevery 1h
FinanceP&L, budgets, anomaly detectionevery 1h
ProductResearch, sourcing, developmentevery 1h
OperationsLogistics, fulfillment, supportevery 1h
DesignBrand, visuals, UXevery 1h
TechShop, integrations, infraevery 1h
LearningCross-agent knowledge transferevery 1h
Cognitive Architecture

Think. Decide. Act. Learn.

Every agent runs BDI cycles — Believe, Desire, Intend, Act, Reflect. Not scripted workflows. Genuine reasoning about priorities, tradeoffs, and next actions.

01
Believe

Read context — KPIs, tasks, events, knowledge base

02
Desire

Identify goals — what needs to happen, what has priority

03
Intend

Plan actions — select tools, set sequence, check autonomy

04
Act

Execute — API calls, content creation, data analysis

05
Reflect

Evaluate — log results, update knowledge, adjust approach

// Agent cycle — every 1–4 hours
async function runCycle(agent) {
// Believe → Desire → Intend → Act → Reflect
const context = await buildContext(agent)
const actions = await decide(context)
await execute(actions, autonomyTier)
await reflect(actions.results)
}
Governance

Four tiers. Full control.

Every tool and action is mapped to an autonomy tier. You configure what agents can do independently and what requires your approval.

T1

Autonomous

Internal analysis, knowledge retrieval, context building. No approval needed.

Read KPIs, search knowledge base, analyze trends

T2

Inform

Actions that change state but are low-risk. Agent executes and notifies you.

Publish content, create tasks, send reports

T3

Approve

High-impact actions. Agent requests approval before executing.

Spend > €200, strategic pivots, external commitments

T4

Human-only

Actions agents cannot take. Escalated with full context.

Legal signatures, physical tasks, account credentials

Self-Improvement

Agents that get better
on their own.

The SICA loop — Suggest, Implement, Check, Accept — lets agents propose changes to their own behavior, test them, and keep what works.

S
Suggest

Agent identifies improvement opportunity

I
Implement

Change deployed as experiment

C
Check

Metrics compared against baseline

A
Accept

Winning changes become permanent

Before

Generic product descriptions, 2% click rate

After

Benefit-focused copy with social proof, 5.1% click rate

+155% improvement — automated
Scale

Multiple ventures.
One system.

Run parallel businesses with isolated budgets, KPIs, and goals. Shared knowledge pool means insights from one venture improve all others.

Isolated budgets

Per-venture spending caps and tracking

Shared knowledge

Cross-venture learning and pattern transfer

Parallel execution

All ventures run simultaneously, 24/7

E-CommerceSaaSAgency

Complexity, handled.

Your team thinks. 42x coordinates.

Built with
TypeScriptClaude (Anthropic)PostgreSQL + pgvectorRedisMCP Protocol
Next

Trust, built in.

Budget governance, audit trails, and human-in-the-loop controls.