ABOUT
CADi keeps a human accountable for the decisions that need one.
As AI moves from advising to deciding, CADi decides, case by case, who decides: the machine alone, the machine with a human, or a human entirely. And it can show why.
AI is moving from advising to deciding. For a decade, models scored and recommended while a person acted. That line is dissolving: agents now pay, refuse, approve, escalate and close. In low-stakes work that is pure progress. In high-consequence work it quietly removes the person who was accountable for the call.
Most systems govern that handoff with a hand-coded threshold: confidence above a line, automate; below it, escalate; in between, shrug. Those rules are fragile, opaque and silent, and they fail exactly when the stakes are highest.
Our mission is to keep a human accountable for the decisions that need one. Not by slowing automation down, but by making the choice between human and machine principled, legible and auditable: decided case by case, on the evidence, and shown.
Everything runs on one deterministic causal engine (“Powered by CADi”): nine layers, a do-calculus core, and a hash-chained audit trail. The engine does not stop at predicting what will happen; it estimates what changed because an action was taken, scores the regret of the permitted alternatives, and records the policy and evidence behind every decision. Its output is an evidence layer we call decision integrity: the proof, per case, of who decided and why they were allowed to.
The range applies that engine to one environment at a time, with more to come:
- CADi PayTHE PAYMENT EVENT
Authorisation, fraud, routing, retry, and the treatment of customers in difficulty.
- CADi ProofEXTERNAL ATTESTATION
Regulatory evidence, any regulator, any jurisdiction.
- CADi PromptTHE PROMPT EVENT
Causal prompt-injection detection for LLM applications.
- CADi AgentTHE AUTONOMOUS AGENT
Causal governance for multi-agent enterprise systems.
- CADi PlayerTHE GAMBLING PLAYER
Player protection and affordability evidence for gambling operators.
- CADi CultureORGANISATIONAL ALIGNMENT
Board-grade evidence that decisions track stated purpose and values.
We run CADi on the same discipline we sell. A human briefs; AI systems produce; enforcement gates check every claim against a register of what we can actually stand behind; a human approves before anything ships. The company is deliberately small and founder-led, built to stay that way: people where accountability lives, machines where production lives.
Two rules are non-negotiable: the human stays accountable for the irreversible act, and when a person is vulnerable, that fact outweighs efficiency, every time, by design.
We say only what is true today and mark designed-but-not-shipped as exactly that. In decision governance, honesty is not a constraint on the message. It is the message.
- Payments first. CADi Pay is the wedge: every transaction is an automated decision with fast, measurable outcomes. Entry is a sidecar: we shadow the live system and measure decision regret before anything of ours touches the decision path.
- A working demonstrator, honestly labelled. On this site the full engine runs live in the browser: an ambient payments simulation running every transaction through the nine-step pipeline, a live causal graph, and expected-loss mathematics operating in a declared synthetic world. It is a prototype and says so; nothing routes on real data.
- The evidence layer. Regulator-readable evidence packs, the monthly regret ledger, and the audit trail that makes every governed decision defensible after the fact.
- The category conversation. A thought-leadership series on why most AI audits ask the wrong question, and why counterfactual reasoning is the only honest way to audit an automated decision.