Product Capabilities
Every request is verified, judged against consent and policy, inspected and recorded as signed, independently verifiable evidence.
PDF · 50 pages · 2.7 MB · October 2026
Download PDFAbout
Consent should be checked where actions happen, not assumed where policies are written.
Mission and vision
Trust should be verified, not assumed.
Mission
qodiqa is open infrastructure that puts a control point in front of an AI provider. Most systems log what an AI did after the fact. qodiqa decides before the call: a request must carry signed, time-limited consent, a defined policy evaluates it, and only then is the provider called.
Every decision, allowed or refused, becomes a signed receipt in a tamper-evident chain, and the intent is written before the provider is reached. If the system fails, it fails closed: no valid consent, no call.
The mission is to make consent, control and accountability enforceable at the moment an AI system acts, not merely promised in a policy or assumed from an earlier approval, and to make the evidence checkable by anyone, offline, without an account and without trusting us.
Vision
The foundations of trustworthy AI should be open, interoperable and accessible. That is why the specification, the gateway and the verifier are published openly, and why public test vectors let independent implementations check themselves against the same results.
Our vision is simple: trust should not depend on promises. It should be built into execution, backed by evidence and open to verification.
- Specify: an open, 32-document standard defining how consent and authorization are evaluated and enforced at runtime (Specification).
- Implement: a reference gateway that enforces authorization before execution and produces cryptographically verifiable evidence (Gateway).
- Verify: independent tools that check evidence against the published protocol (Verifier).
Principles
Six principles of the standard.
Deterministic enforcement
Consent evaluation yields a binary, reproducible outcome on every execution. No probabilistic fallback.
Runtime interception
Enforcement is applied at the inference boundary, before any output is generated or any downstream action is taken.
Refusal by default
If consent cannot be verified, execution halts. The system does not degrade to permissive operation.
Cryptographic auditability
Each decision is cryptographically signed and independently verifiable outside the enforcing system.
Scope-bounded consent
Consent is evaluated against a defined execution scope. Operations outside that scope are blocked.
Consent artifact immutability
A consent artifact cannot be modified once issued and recorded. Any change requires a new artifact with its own identifier.
Standard and implementation
One standard. Independent implementations.
Anyone can build an implementation. Anyone can verify its evidence.
Specification
Defines what enforcement must do.
Read the specificationGateway
The reference implementation, with its capabilities and limits documented openly.
See the gatewayVerifier
Checks evidence against the published protocol, independent of the gateway’s code.
Open the verifierThe rules are open, enforcement is explicit, and the evidence can be independently verified.
Founder
A message from the founder.
AI must not be left uncontrolled. And control that comes after execution may come too late.
I founded qodiqa on a simple conviction: as AI becomes capable of making decisions, accessing systems and executing actions on its own, trust alone is no longer enough. We cannot build the future of AI on policies that exist only on paper, safeguards that can be bypassed, or audit trails that explain what happened only after the consequences have occurred. Control must exist where execution happens, before an action is allowed to proceed.
My vision is a future in which AI systems are not merely expected to follow rules but are technically constrained by them: every controlled action has a defined authorization, every decision leaves verifiable evidence, and verification does not depend on trusting the system being examined. I do not want to slow AI down. I want its progress to last, and I want trust to be something that can be tested, not another promise.
“Rather than debating AI’s technical limitations, we must examine what it was empowered to do, who authorized it, when, and whether any third party possesses the evidentiary basis to confirm it.”
Work with qodiqa: see Careers.