AI Agent Control Plane
An AI agent control plane is the governing layer over a fleet of autonomous agents — where authority is granted, scoped, and revoked; where every consequential action is logged; and where a thing is only “done” when reality confirms it. NXTG.AI runs one internally and measures it in public: 570 agent-wakes are logged in an openly downloadable dataset, and the control plane itself is documented in a pre-registered origin report.
A control plane spans layers: the platform that governs, the arm that governs delivery, and the check that governs evidence. Each renders its verbatim public-catalog description. Dx³ is internal by design; its receipts are public.
The data-intelligence platform we run ourselves. PostgreSQL, pgvector, and Apache AGE unified. No public URL by design; the published research is the evidence.
Governance-native AI development: drift detection, CRUCIBLE auditing, and security guards for Claude Code, Codex CLI, and Gemini CLI.
AI claim verification CLI: paste what an AI told you; it extracts every claim, verifies against live sources, and scores the risk.
Four manifesto principles define what a control plane must guarantee. They render verbatim from the site manifesto.
03 / Governance is infrastructure
Policy-as-code, model registries, scoped authority, attestation pipelines. Auditable by default. Every exception visible.
04 / Everything consequential is traceable
Every consequential action and artifact is content-addressed, logged, and observable end to end. If it happened, you can prove what happened, why, and under whose authority.
11 / You authorize what acts in your name
Nothing acts as you without explicit, inspectable, revocable authority. Inside that scope the system acts without asking again; outside it, it stops or escalates. What you see is what you authorize.
12 / Reality outranks status
A thing is done when evidence from the target environment confirms it — not when a task, a dashboard, or a deployment says so. We verify against reality, never against our own claims.
Every claim on this page traces to a publicly probeable surface. These are the links.
What is an AI agent control plane?
An AI agent control plane is the governing layer over a fleet of autonomous agents. It is where authority is granted, scoped, and revoked; where every consequential action is logged and traceable; and where a task is only done when evidence from the real environment confirms it, not when a dashboard says so.
Does NXTG.AI run an AI agent control plane?
Yes, internally. NXTG.AI runs its own multi-agent control plane and measures it in public: 570 agent-wakes are logged in an openly downloadable dataset, and the platform is documented in a pre-registered origin report with data deposited on Zenodo. The platform itself, Dx3, is internal by design; the receipts are public.
How do you prove a control plane actually works?
By refusing to take its own word for it. NXTG.AI pre-registers predictions before the outcomes are known, deposits the underlying data as openly downloadable datasets, and treats verified reality as outranking any status field. The standing rule is that a thing is done when evidence from the target environment confirms it.