CTROLL — Control plane over AI agents

The problem

The most valuable AI in the world is the AI banks can’t deploy

Every regulated firm wants agents. Almost none can turn one loose — the law demands human oversight, independent verification, and an unbroken record of what happened.

Standard agent platforms offer none of it. So the firms with the most to gain watch demos they can never ship.

CTROLL exists to end that

One control plane · Three things an agent framework does not own

Separate reviewerWorker and reviewer must be different providers. A self-review cannot be recorded.
Spend quotesPlanning and execution each need an owner-bound quote. No provider call starts without one.
Verifiable evidenceHash-chained mission log, verify endpoint, DORA register and incident export.

Runtime contract

What the plane actually owns

RolesCEO plans. One worker runs. One reviewer from another provider checks. That is the runtime, not a seven-agent roster.
StatePostgreSQL with FORCE row-level security. Providers are adapters. They do not own the mission graph.
ProofVerify the hash chain on the mission. Export the DORA register and incident file from the same ledger.

How it works

Instruct · Govern · Prove

Three stages, and a gate before anything irreversible

Delegation: six signals converge into one organised network.

01
You give one instruction

“Analyze these three markets and rank them”
02
CTROLL runs it under control

An AI lead plans and delegates to named agents. A second model reviews each result. Every spend and decision hits a gate before it happens.
03
You get the result — and the proof

The output, the review verdict, and a step-by-step evidence trail with cost and provenance. Ready for the compliance officer, not just the CEO.

Evidence

Other products show you a chat · We show you the receipt

This is the record shape the control plane writes for every mission: the task graph, who reviewed whom, how many attempts it took, and which requirement each piece of work is answerable to. Run the demo below and yours looks like this.

A second agent, from a different provider, rejects the first attempt. The reviewer is never the implementer.

demo passed

Demonstrate a bounded launch-readiness control flow

a4f19c2b07d3 · AUTO · PLAN R1

$0.00 · 0 provider calls

Bounded task graph

Primary implementation accepted · 2 reviews

2 ATTEMPTS

Attempt 1 rejected Planted demo defect detected. Resolve and resubmit evidence.
Attempt 2 accepted Independent review passed on resubmitted evidence.

Independent risk and continuity analysis accepted · 1 review

1 ATTEMPT

Attempt 1 accepted Independent evidence and unresolved risks returned.

The reviewer is never the implementer. The orchestrator refuses a verdict whose author matches the worker, so a self-review cannot be recorded as one.

Requirement-to-evidence trace

BR-012 evidencepassed

BR-022 evidencepassed

BR-042 evidencepassed

BR-062 evidencepassed

ARC-022 evidencepassed

GOV-032 evidencepassed

Recorded events

job submitted plan created dispatch started task started task reviewed rework requested task started task reviewed job finalized

For regulators

Built to clear the bar that stops ungoverned AI

CTROLL was engineered inside a licensed EU credit firm that legally could not use uncontrolled AI. Every control a regulator asks for is in the architecture — not bolted on after.

Human oversight before consequential actions

EU AI Act Art 14 · GDPR Art 22
Approval gates on spend, deploy and decisions

Automatic logging of every event

EU AI Act Art 12
Per-task evidence log with model, cost and state

Independent verification of outputs

EU AI Act Art 9
A second model reviews every result

Third-party provenance and portability

DORA
Provider-neutral routing with a source trail

Explainable decisions

GDPR Art 22 · EBA
Decision → evidence → source, on every step

Engineered for regulatory approval

Designed to pass the audit, not just the demo

A five million share order, halted at the gate: nothing irreversible happens until a person signs.

Why now

Regulation just made this mandatory

High-risk AI in finance now requires logging, human oversight, and traceability — by law. The EU AI Act’s high-risk obligations phase in through 2027. DORA has governed financial ICT since 2025. The SEC already enforces WORM recordkeeping and prosecutes AI-washing. The firms that can prove control will deploy. The ones that can’t, won’t.

Forged inside a licensed lender, not a lab: CTROLL wasn’t dreamed up to chase a trend. It was built inside an EU-authorized credit firm, against a real regulatory bar, from day one. Most “AI governance” is a slide deck. Ours runs governed multi-agent missions in production today — planned, delegated, independently reviewed, and logged with cost and provenance on every step.

Shipping next, and aimed at your auditor: tamper-evident WORM archival with cryptographically chained evidence, exportable to non-rewriteable storage for SEC 17a-4; DORA reporting exports with the third-party register and incident log in reportable form; SOC 2 and ISAE 3402 attestation of the control system; and AI Act conformity assessment for high-risk deployments. We ship the code in weeks. We pursue the attestations in parallel.

What shipping next looks like: a sealed evidence log, handed over in a form an auditor can read. Tamper-evident archival is on the roadmap, not yet in place.

Owner control plane

Command the agents
Keep control

Turn one outcome into a bounded plan, a separate review gate, one repair and traceable control evidence. The surface below is the live control plane, not a screenshot.

The owner releases it. Control is not the power to say no — it is the cover to say yes.

DEMO DEFAULT SCOPE — SIGN IN FOR LIVE STATUS

CT / OPS

CONTROL LEDGER CONNECTED

CURRENT VIEWMISSIONS

PROVIDERSCHECKING AGENTSIDLE RESEARCHDISABLED LEDGERCONNECTED

01

Issue mission

01 / EXECUTION

Free demo uses simulated responses. No paid model calls.

DEMO ROUTEReady · no provider calls
OWNER APPROVALNot required

EXECUTION ENVELOPEOWNER BOUNDED

BACKENDCHECKING

DATABASECHECKING

WORKERCHECKING

PROVIDERCHECKING

PUBLICATIONCHECKING

Checking the production runtime boundary.

Reading current server-owned readiness evidence…

MODE: AUTO
ROUTE: Demo / local
WORKFLOW: Bounded demo
CALL CAP: 0 paid
BUDGET: $0.00
REPAIR: 1 cycle max
CONCURRENCY: 2 workers max
RECURSION: Disabled
RESEARCH: Disabled
SEARCH EST.: No research spend

WORKERREVIEWEROWNER

Provider identities and the exact spending envelope are bound into each live approval quote before execution.

Source research is disabled by default. Model answers use only supplied context unless this service reports research ready.

MISSIONS0durable records

IN FLIGHT0nonterminal missions

REVIEW PASSES0demo and live shown separately

PROVIDER SPEND—waiting for session

LATEST LIVE MISSION / READ ONLY

Cost and budget

EST. / COMMITTED: —
OPEN RESERVATIONS: —
OBSERVED BILLING: —
CALL CAP: —
REMAINING BUDGET: —

No live mission budget is loaded. Estimates and observed billing are never presented as the same value.

CONTROL-PATH DEMO Uses simulated task and reviewer responses. Select a configured provider for paid text work.

No missions yet. Give CTROLL one outcome above.

The AI agent control systemfor finance and regulated markets

The most valuable AI in the world is the AI banks can’t deploy

What the plane actually owns

Instruct · Govern · Prove

You give one instruction

CTROLL runs it under control

You get the result — and the proof