cmpliance is the agentic workforce for compliance. Agents screen, classify, gather evidence, and assemble the evidence pack. Humans review, approve, and sign off.
The shift
Signals, evidence, and deadlines arrive in pieces. cmpliance turns each into owned work with its evidence and proof trail.
Agentic resolution
Owned work with reviewer gatesOwned work
Audit trails built for reviewEvidence pack
Evidence packs you can export and defendObligation
Why agentic
cmpliance agents screen, read, map, and classify. They attach the evidence and hand the completed work to your officer for review and decision.
DecideYour callM. Fischer
One case. Agents work. Traceable evidence. Your call.
Brenvik Components GmbHEU / CH · KYC/AML onboarding
MakerA. KellerCheckerM. Fischer
Screening agentClassification agentEvidence agentNarrative agent
HighM. FischerNominee for a sanctioned individual
Your callApproveFlagFlagged · escalated to MLRO
Evidence packExport PDF
EU AI Act review
Pick up where you left off. Queue sorted by urgency.
Corvenna Risk Systems GmbHEU AI Act
#2risk score 72DecidedL. Brandt
#1risk score 55DecidedL. Brandt
What name matching misses
A name‑match system runs a counterparty against sanctions and watch lists and returns a result. cmpliance reasons over the structure between entities.
Reasoning depth
Name match1 hop · direct lookup
cmpliance3 hops · ownership chain
Signals checked
Name matchName string · fuzzy match
cmplianceOwnership · cross‑jurisdiction directorships · adverse media
Sources
Name matchOFAC · EU
cmplianceCorporate registers · sanctions lists · media reports
Evidence pack
Name matchNone · result log only
cmplianceEP-2026-4182 · sha256:4e1d…a86b
Name matching finds a name. cmpliance finds the structure.
Architecture
cmpliance combines language‑model reasoning over documents with graph reasoning over entity structures, in a shared embedding space. Frontier models are interchangeable. The structural reasoning layer is cmpliance’s own. Your data and your calibration stay in your tenant.
Each layer is one case. The structure stays.
Virtual data room
Share evidence packs across reviewers and auditors.
Verification
Audit ledgerthis pack
A pack’s hash is written to the audit ledger, so a copy can be checked against it.
Calibration
cmpliance calibrates to your organization’s evidence standards, escalation thresholds, jurisdictions, and review decisions. Approvals, rejections, overrides, and their rationale adjust how agents assess and escalate cases for your team, starting with the next assessment. It does not train a generic vendor model on your data.
Evidence packEP-2026-3907Decided
Domestic subsidiary · single‑layer ownership · CH
Risk ratingHigh Medium
Decided by M. Fischer · 7 May 2026, 10:14 UTC
Calibration entryCAL-2026-0391Decided
OverrideRisk rating: High → Medium
Scopetenant: pilot‑ch · scope: kyc‑onboarding · effective: next run
Hashsha256:b41c…0e7d
TrainingNo. Signal scoped to tenant, not used to train public or foundation AI models.
hash‑chained
CaseCASE-2026-KYC-3E1CAwaiting decision
Domestic subsidiary · single‑layer ownership · CH
Proposed ratingMedium
PrecedentM. Fischer’s override CAL-2026-0391
cmpliance is the agentic workforce for compliance. Agents screen, classify, gather evidence, and assemble the evidence pack. Humans review, approve, and sign off.
The shift
Signals, evidence, and deadlines arrive in pieces. cmpliance turns each into owned work with its evidence and proof trail.
Why agentic
cmpliance agents screen, read, map, and classify. They attach the evidence and hand the completed work to your officer for review and decision.
How it works
One case. Agents work. Traceable evidence. Your call.
EU AI Act review
A high‑risk AI system influencing credit decisions lacks bias‑testing and human‑oversight evidence; escalate to the compliance lead before production use.
Reviewer action
Request additional vendor evidence before approval; escalate to compliance lead because the system affects credit decisions for natural persons.
Corvenna Risk Systems GmbH · Rationale. The audit‑log sample shows credit scores overridden without a recorded reviewer rationale; insufficient logging and human‑oversight evidence under EU AI Act Articles 12 and 14.
What name matching misses
A name‑match system runs a counterparty against sanctions and watch lists and returns a result. cmpliance reasons over the structure between entities.
Adverse mediaNot checked2 reports · naming the nominee director
No match on counterparty name
Nominee for a sanctioned individual
Escalated to MLROName matching finds a name. cmpliance finds the structure.
Architecture
cmpliance combines language‑model reasoning over documents with graph reasoning over entity structures, in a shared embedding space. Frontier models are interchangeable. The structural reasoning layer is cmpliance’s own. Your data and your calibration stay in your tenant.
Each layer is one case. The structure stays.
Virtual data room
Share evidence packs across reviewers and auditors.
Calibration
cmpliance calibrates to your organization’s evidence standards, escalation thresholds, jurisdictions, and review decisions. Approvals, rejections, overrides, and their rationale adjust how agents assess and escalate cases for your team, starting with the next assessment. It does not train a generic vendor model on your data.
1Reviewer decision
3Next assessment uses your calibration
2Labeled signal · for policy calibration
Compounding calibration
Every month of reviewer decisions calibrates your agents further to your thresholds and precedents. It stays in your tenant, and the sooner you start, the sooner it compounds.
Agentic workforce
Agents prepare the repetitive work, preserve the proof trail, and hand decisions back to compliance teams.
Submit a counterparty. Agents screen, read, map ownership and classify risk, then assemble a complete evidence pack. You review it and make the call.
Signals become assigned work, evidence checkpoints, reviewer gates, and decision records. Every step is logged, so the evidence pack shows not just the result, but how the decision was reached.
Agents classify your AI system’s risk level, prepare the Annex IV points they can, list what is missing, and assemble an evidence pack with regulatory references for reviewer approval.
Submit a campaign asset and how it was made. Agents compare versions, check which disclosure rules apply to the market and the channel, list the evidence that is missing and prepare a recommendation. Your reviewer makes the call, and the decision stays bound to that exact version.
Pricing
Agents do the jobs: they screen parties, read registry extracts into cited facts, map ownership, classify risk and re‑check entities over time. Everyone on your team can use cmpliance, and the decisions stay with your officers.
One month, side by side. Over twenty working days 288 jobs arrive, more each week, and all of them are done. A reviewer joins, then another reviewer. Seat‑based pricing counts the people: 15 paid seats, then 16, then 17. cmpliance counts the jobs agents finish: 288 jobs finished; everyone on your team can use it, with no seat fees.
Track record
paterhn has built AI applications for RegTech, FinTech, and other regulated industries since then. Three of those deployments:
Transaction screening
Agentic screening across the compliance graph.
Private banking
Regulatory documentation prepared end‑to‑end for compliance team review.
Evidence infrastructure
Deterministic evidence packs with hash‑anchored audit trails.
Selected predecessor‑system deployments in regulated industries. Details available under NDA. These are not cmpliance production results.
Insights
Evidence‑led notes on audit readiness, control design, and keeping accountable humans in control of agentic compliance.
Approval is a point in time. When a new sanctions signal arrives for an approved customer, a reassessment opens against the approved baseline, the comparison shows what changed, and an analyst decides what the change means. Follow that path in cmpliance.
An AI label does not explain a publication decision. Companies need the asset's history, the applicable rules and an accountable reviewer tied to the version that goes live.
An approved status tells a reviewer what happened. Explaining why requires the source evidence, human rationale and recorded actions that supported the decision. This cmpliance walkthrough follows that path.
cmpliance.ai is the official website for cmpliance, an evidence‑led compliance product operated by paterhn GmbH in Switzerland.
Core production architecture is designed with Switzerland and EEA-oriented processing in mind where practical. Specific provider locations and transfer safeguards are governed by the applicable customer agreement and subprocessor terms. See Security.
Our initial scope is KYC/AML onboarding and EU AI Act risk assessment. The product is designed to bring evidence, findings and a human decision into one review record. A workflow or evidence pack does not by itself establish compliance with a regulation. Contact us to confirm the supported scope for your case.
No. We do not use Customer Data to train public or foundation AI models. Customer Data is processed to provide the Service and as otherwise permitted by the applicable customer agreement.
Evidence and document workflows are designed to support integrity checks, auditability, and source traceability. Compliance workflows are designed to record user actions, system actions, evidence events, and decision history. Specific controls may vary by plan, feature, deployment model, and signed customer agreement. See Security.
cmpliance.ai is built by paterhn.ai, a Swiss AI development studio with production AI systems in operation since 2001, multi‑agent systems running in enterprise environments since 2017, and transformer‑era agent work in production since 2019.