Who decided?
Praefex is designed to require governed validation before decisions enter the committed record, giving reviewers a clearer path from output to accountable action.
AI governance and efficiency infrastructure
AI decisions you can verify.
Praefex is a proof-backed AI governance and routing system focused on reliability under load, tamper-evident receipts, and measured claims. It helps enterprise governance, security, risk, compliance, and AI infrastructure teams preserve evidence, govern decisions, and reduce unsupported AI action.
Enterprise AI has four questions it usually cannot answer.
Most AI deployments can generate answers. Fewer can prove what happened, preserve decision context across sessions, control unverified outputs, or reduce waste without quietly degrading quality.
What evidence supported it?
Tamper-evident receipts preserve the decision trail so reviewers can inspect what was accepted, why it was accepted, and which checks were applied.
What changed?
Persistent operational memory helps AI work carry verified context forward instead of restarting from an invisible blank slate every session.
What was blocked?
Unsupported decisions are designed to stay out of the governed record; consensus reduces single-node failure and manipulation risk.
One layer. Three jobs: govern, secure, optimize.
Praefex is an AI assurance layer for enterprise AI governance and regulated AI workflows: decision control, evidence preservation, and quality-gated compute routing in one operating posture.
Decisions need memory and accountability.
Praefex preserves decision context across sessions so AI work does not restart from an invisible blank slate. Policy gates are designed to evaluate decisions before they enter the governed record.
Tamper-evident records, not blind trust.
Praefex records decisions in a tamper-evident chain and uses governed validation to reduce unsupported commitment risk. The goal is to prevent unverified outputs from silently becoming official.
Spend compute where quality requires it.
Praefex is being validated as a quality-gated routing layer: simpler work can use lighter paths, harder work can escalate, and outputs remain subject to correctness gates before commitment.
Built as an operator layer, not a chatbot.
PICA is the Praefex Integrated Control Arm: a reference operator surface for configuring verification depth, containment posture, consensus policy, routing behavior, and audit-chain controls across governed AI workflows.
Governance teams need controls they can inspect.
PICA presents the Praefex governance layer as policy, validation, routing, audit, and recovery controls rather than a one-off prompt interface.
PICA shows the operating model.
PICA is a reference control plane for smaller deployments and technical review. Enterprise deployments can integrate the same governance layer through private services, APIs, audit exports, and customer-owned dashboards.
Public materials show shape, not mechanism.
Architecture notes, validation receipts, test results, and methodology are available to qualified reviewers under NDA. Security-sensitive implementation details are handled through controlled disclosure.
Measured internally, scoped honestly.
Praefex has internal validation evidence for governed routing and GPU-energy reduction, with methodology, conditions, receipts, and limits available to technical reviewers under NDA. Public claims stay scoped until evidence packets and release verification are approved.
Quality-gated routing validation
In validation cycles where the 27B baseline was healthy, the codeaware policy typically used about 37-39% less compute-proxy at identical 18/18 correctness. This is an internal 18-task benchmark using compute-proxy, not measured production energy. Cycles where the 27B baseline was unhealthy are excluded from this comparison.
Scoped energy benchmark
A scoped internal GPU-energy benchmark is available with methodology, raw artifacts, and replication status for qualified reviewers.
Receipts and methodology
Measurement records, hardware notes, task conditions, and known limitations are available for evaluator review under NDA.
For technical reviewers.
Technical review is available under NDA. Praefex is built for high-trust environments, so public materials describe what the system verifies, not every mechanism by which it verifies it.
Architecture notes
Governed validation, persistent operational memory, quality-gated routing, audit-chain controls, and integration patterns for enterprise AI workflows.
Validation receipts
Scoped methodology, measurement records, evaluation conditions, limitations, and replication posture for qualified technical, investor, or enterprise review.
Disclosure boundary
Certain implementation details are withheld from public materials because they describe security-relevant controls and recovery behavior.
AI Governance FAQ
What is AI decision governance?
AI decision governance is the control layer that determines whether an AI output can become an accepted decision. It covers validation, provenance, audit trails, escalation, and blocking unsupported outputs before they commit.
What does Praefex mean by verifiable AI decisions?
A verifiable AI decision is a decision with a preserved record of what was decided, why it was accepted, what evidence supported it, and which checks were applied before commitment.
What is a tamper-evident AI audit trail?
A tamper-evident AI audit trail is a record where later alteration is detectable. Praefex uses this framing deliberately: it does not claim alteration is impossible or permanent.
Does Praefex prove every AI answer is true?
No. Consensus and validation reduce single-node failure and manipulation risk, but they are not proof that every statement is factually correct. Praefex is designed to block unverified decisions from committing and preserve evidence for review.
How does Praefex help with AI compute efficiency?
Praefex is validating quality-gated routing that sends work to the right model path while preserving answer standards. Efficiency claims must state the benchmark, baseline, and measurement type.
Can technical reviewers inspect the methodology?
Yes. Praefex can provide methodology, receipts, and scoped validation materials under NDA for qualified technical, investor, or enterprise evaluation.
Request an evaluation.
Praefex is in active internal deployment and pre-commercial enterprise evaluation. We are opening a limited number of conversations with security, compliance, governance, and infrastructure teams evaluating AI decision control.