AlienCortex.com Alien intelligence. Human control. Download HTML · Download PDF · Build log · Back to the site
How it's built · Part 1 of 2 · Architecture

How Alien Cortex is built

The full aBOS map: the front door, the factory, the six core agent classes, the safety rules and the patent-pending mechanisms. Part 2, the build log, is below. You can also download this map as HTML or as a PDF.

The aBOS loopsEnd-to-end flow: people reach the AI Audit, the Cell Factory builds each company a cell where six agent classes work, and anything irreversible passes a named-person approval gate before reaching the outside world. Four loops sit on this flow. 1 Governance loop, at the factory and the gate: rules compiled in at creation, agents propose, a named person approves the irreversible, trust rises or falls with the override rate. 2 Self-learning loop, in the cell and at the gate: agent drafts, a person approves or corrects, the correction is captured, memory is updated by sensitivity tier. 3 Self-improving loop, in the cell: propose a better way to work, test only rules people chose, adopt or reject with the reason kept, revisit when the reason stops being true. 4 Compounding loop, at the factory and the shared library: a pattern proven in one cell passes a provenance gate with no customer data, joins the shared library, and the factory starts new agents and cells from it. Only patterns go up, never data.HOW THE SYSTEM RUNS, END TO ENDPeopleowner, team, clientsAI Auditthe one front doorCell Factorybuilds every cell14Company's cellsix agent classes at work23Approval gatea named person says yes12Outside worldsends, payments, filingsNumbered badges show where each loop sits on the flow.1Governance loopRules compiledin at creationAgent proposesNamed personapproves theirreversibleTrust rises orfalls withoverride rateRules are built into each cell when it iscreated, not set later. Agents propose. Anamed person approves anything irreversible.An agent earns more freedom only with a low,falling correction rate and passing tests.2Self-learning loopAgent draftsPerson approvesor correctsCorrectioncaptured at thatmomentMemory updated,by sensitivitytierEvery correction is captured when it happens.The override rate, the share of work a personhad to change, shows whether the system islearning this business. Finance data neverenters general memory.3Self-improving loopPropose a betterway to workTest only rulespeople choseAdopt, or rejectwith the reasonkeptRevisit when areason stopsbeing trueThe system improves how the work is done, notonly the work. It never tests changes tolaws, contracts or safety limits, and itkeeps every rejected idea with its reason.4Compounding loopPattern provenin one cellProvenance gate:no customer dataShared libraryFactory startsnew agents andcells from itWhat one company's agents learn about a classof job goes to the shared library, and everyother company benefits. Only patterns cross,never data. If ancestry cannot be proven, itdoes not cross.override rate sets trustcaptured correctionsproven patternsHOW IT GROWSThe factory keeps stamping cells. The shared library keeps compounding patterns.Cell Factory: onecore release plus anexact pack lockstampsCellCellCellCellCellNext cellonly patterns go up, never customer dataShared library of proven patternsnew cells start fromproven patternsNew agents and new cells start from patterns already proven elsewhere, so improvements compound instead of adding up one company at a time.The aBOS loopsEnd-to-end flow: people reach the AI Audit, the Cell Factory builds each company a cell where six agent classes work, and anything irreversible passes a named-person approval gate before reaching the outside world. Four loops sit on this flow. 1 Governance loop, at the factory and the gate: rules compiled in at creation, agents propose, a named person approves the irreversible, trust rises or falls with the override rate. 2 Self-learning loop, in the cell and at the gate: agent drafts, a person approves or corrects, the correction is captured, memory is updated by sensitivity tier. 3 Self-improving loop, in the cell: propose a better way to work, test only rules people chose, adopt or reject with the reason kept, revisit when the reason stops being true. 4 Compounding loop, at the factory and the shared library: a pattern proven in one cell passes a provenance gate with no customer data, joins the shared library, and the factory starts new agents and cells from it. Only patterns go up, never data.HOW THE SYSTEM RUNS, END TO ENDPeopleowner, team, clientsAI Auditthe one front doorCell Factorybuilds every cell14Company's cellsix agent classes at work23Approval gatea named person says yes12Outside worldsends, payments, filingsNumbered badges show where each loop sits on the flow.1Governance loopRules compiled inat creationAgent proposesNamed personapproves theirreversibleTrust rises orfalls withoverride rateRules are built into each cell when it is created, not setlater. Agents propose. A named person approves anythingirreversible. An agent earns more freedom only with a low,falling correction rate and passing tests.2Self-learning loopAgent draftsPerson approves orcorrectsCorrectioncaptured at thatmomentMemory updated, bysensitivity tierEvery correction is captured when it happens. The overriderate, the share of work a person had to change, shows whetherthe system is learning this business. Finance data neverenters general memory.override rate sets trust (loop1), captured corrections feedloop 33Self-improving loopPropose a betterway to workTest only rulespeople choseAdopt, or rejectwith the reasonkeptRevisit when areason stops beingtrueThe system improves how the work is done, not only the work.It never tests changes to laws, contracts or safety limits,and it keeps every rejected idea with its reason.proven patterns feed loop 44Compounding loopPattern proven inone cellProvenance gate:no customer dataShared libraryFactory starts newagents and cellsfrom itWhat one company's agents learn about a class of job goes tothe shared library, and every other company benefits. Onlypatterns cross, never data. If ancestry cannot be proven, itdoes not cross.HOW IT GROWSThe factory keeps stamping cells. The shared librarykeeps compounding patterns.Cell Factory: one core release plus an exact pack lockCellCellCellNext cellonly patterns go up, never customer dataShared library of proven patternsNew agents and new cells start from patterns already provenelsewhere, so improvements compound instead of adding up one companyat a time.
The aBOS flow and its four loops. The interactive map below has the detail for each part.
aBOS System Explorer
aBOS by AlienCortex · 18 October 2026

Your own AI, as powerful as a company.

Imagine a toy factory that builds a little team of helper robots for each company. The robots do the busy work, a grown-up always says yes before anything important happens, and the company can take its robots home whenever it wants.

aBOS gives each company its own private AI team. One factory builds every team from the same tested design, so each new company is cheaper and safer to set up. The AI drafts and prepares. People decide. The company can leave with everything at any time.

aBOS is a governed, multi-tenant agent operating system. A Cell Factory compiles one versioned core plus a signed pack lock into an isolated per-tenant cell, with enforcement metadata bound at instantiation. Six mandatory agent classes run under permission envelopes, sensitivity-tiered memory and named-human gates; Finance inference is local-only.

1factory, one tested design for every cell
6agent classes in every core
8provisional patent applications filed (patent pending)
0payments, sends or deletions without a named person
The whole picture

One map, every part

Tap any box to learn what it does. Use the buttons at the top to hide or show parts of the picture.

Tap any part to see what it does, which patent-pending mechanism protects it and how it fits ExO 3.0. Use the layer buttons to show only what you are explaining.

Select a component for its contract, protecting filings, ExO 3.0 mapping and moat class. Layer toggles filter zones.

People, front door and the outside world

who uses it and what it talks to
▼ the AI Audit decides the journey · the factory builds the cell ▼

The Cell Factory

builds, upgrades and removes every cell
▼ one core release + an exact pack lock + company settings + the client's persona ▼

A cell: one company's private aBOS

example: the AlienCortex cell
Core · in every cell, never optional · six agent classes
Core · platform services
Packs · chosen per company · add skills, never authority

Client cell: a services firm

Same core: six classes, console, mail, brain, local Finance, governance, export
funnelsales-marketingchat-surface

Client cell: a construction company

Same core, same rules
construction 1.0ezra-calendar

Client cell: real estate and development

Same core, same rules
real estate (planned)investor-pipelineconstruction
▲ only patterns go up, never customer data · only a content-free heartbeat goes out ▲

Shared across all cells

what the network learns, and how it is watched
For a client meeting

Guided tour, nine stops

Step through the story in order. Each stop selects the part on the map above and explains it at your chosen reading level.

1

Intellectual property

Eight patent-pending mechanisms, and where each one lives

These are eight clever ideas we wrote down and sent to the patent office first, so they count as ours while the office looks at them.

Eight provisional patent applications are filed with the USPTO. Each protects a mechanism that runs inside aBOS. Tap "Show on map" to light up the parts it covers.

Eight US provisional applications under 35 USC 111(b), filed March to October 2026, each standalone (no benefit claims). Scope is described by subject matter, never by claim number.

Say it correctly: "eight provisional patent applications filed" or "patent pending". These are not granted patents and have not been examined.
Method

Where ExO 3.0 fits

ExO 3.0 is a recipe, written by Salim Ismail, for how a company should work when AI helpers do lots of the jobs. aBOS is the kitchen that actually cooks that recipe every day.

ExO 3.0 is Salim Ismail's framework for the AI-native company (The Organizational Singularity). It describes the destination. aBOS is running software that gets a company there: every ExO 3.0 component maps to real parts of the system. The framework is credited to its authors; the software that implements it, and its eight filed mechanisms, belong to AICSAI LLC.

ExO 3.0 = MTP-as-protocol (Constraint, Decision, Identity layers) + DRIVE (5) + SHAPE (5) over the Intelligence Stack (Four Pillars: trusted evals, searchable logs, granular rollback, human review queue; GOVERN/ASSURE). aBOS implements each element in its data plane. Select a letter to highlight its implementation on the map.

MTP as a protocol

DRIVE · the intelligence engine

SHAPE · the organizational form

Intelligence Stack · the engine block

ExO 3.0 in one picture

Engine, drivetrain, chassis

Intelligence StackThe engine block. In aBOS: the core in every cell.
DRIVEThe drivetrain: speed and learning. In aBOS: tiers, closed loop, packs, autonomy ladder.
SHAPEThe chassis and brakes. In aBOS: hard rules, named humans, local Finance, exit right.

Select any letter on the left to see how aBOS implements it, then look at the map to see the parts light up.

"DRIVE without SHAPE is a fuse waiting for a spark." aBOS builds both: speed only inside rules that cannot be switched off.
Moat

How much of it is ours

We count every part of the picture and colour it by who made it. Most of the important parts are our own ideas.

Every part on the map is counted below by what protects it. The counts are computed from the map itself, so they change if the map changes. Parts that third parties make, like open-source models, are credited and not claimed.

Classification per component: covered by at least one filed mechanism (P1 to P8); otherwise trade secret / know-how; otherwise credited third-party or open source. Counts are live over the component dataset.

1 · Filed mechanisms (patent pending)

Eight provisional applications cover the cell model and local Finance (P1), safe learning, memory tiers and earned autonomy (P2), governance built in at creation (P3), workflow redesign (P4), multi-person discovery (P5), personas (P6, P7) and the AI meeting participant (P8).

2 · Trade secret and know-how

The factory implementation, the single cell definition, golden test sets, attempted-violation test suites, release and recovery machinery, and the operating discipline that runs it.

3 · Compounding assets

The shared library of proven patterns grows with every cell (only patterns cross, never data). Each client's persona relationship is licensed, not exported. Intelligence density rises as override rates fall.

4 · Credited, not claimed

ExO 3.0 (Salim Ismail and contributors), open-source parts (local models, vector store, web framework, agent runtime, Postiz as a separate process) and cloud model vendors for non-Finance work.

Promises that cannot be switched off

The hard rules

A rule always beats a priority. These are enforced underneath the buttons, in the data plane, and each one has a test that tries to break it.

Words

Glossary

How it's built · Part 2 of 2 · Updates and journey

Build log

Build log entries start here soon. We will post short, dated notes on what changed in aBOS and why, and a timeline of how Alien Cortex got here.