Connectors
Use a connector when the governed surface is a vendor platform you cannot instrument directly inside your own runtime.
The connector surface is a separate package: pip install primust-connectors. Source lives at github.com/primust-dev/primust-connectors.
What You Can Actually Use Today
The current package exposes eight importable Python connectors. If you came here expecting a concrete list, this is the list.
pip install primust-connectors
| Connector | Use case | Status | Typical ceiling |
|---|---|---|---|
ComplyAdvantageConnector | AML entity screening | Python package | Attested |
NiceActimizeConnector | AML monitoring and SAR review | Python package | Witnessed / Attested |
FicoBlazeConnector | Credit decisioning BRMS | Python package | Execution today, stronger with in-process path |
IBMODMConnector | Enterprise BRMS and underwriting | Python package | Execution today, stronger with in-process path |
FicoFalconConnector | Fraud detection and thresholds | Python package | Attested / partial Execution |
PegaDecisioningConnector | Regulated next-best-action | Python package | Attested |
UpToDateConnector | Clinical decision support | Python package | Execution today, stronger on bounded stages |
GuidewireClaimCenterConnector | P&C claims adjudication | Python package | Attested / boundary evidence |
Reference Or Spec-Only Connectors
There are also reference connectors in the repo that are not importable Python package classes today. These are useful for enterprise or partner-specific paths, but they should not be documented as self-serve Python connectors.
| Platform | Status | Notes |
|---|---|---|
| InterSystems HealthShare | Java spec | Requires separate Java SDK integration. |
| Sapiens DECISION | Java spec | Reference adapter, not importable from the Python package. |
| Duck Creek Technologies | C# spec | Reference extension, not a Python connector. |
| Majesco CloudInsurer | C# spec | Reference extension, not a Python connector. |
| Sapiens ALIS | C# spec | Reference extension, not a Python connector. |
Quick Example
Pythonfrom primust_connectors import ComplyAdvantageConnector
connector = ComplyAdvantageConnector(
ca_api_key="ca_live_...",
primust_api_key="pk_live_...",
)
connector.register_manifests()
p = connector.new_pipeline()
result = connector.screen_entity(
pipeline=p,
entity_name="Acme Corp",
entity_data={"name": "Acme Corp", "country": "US"},
)
Each connector records evidence into the same Primust governance chain. The difference is that the evidence is captured at the vendor boundary rather than inside your own process.
Connector Notes By Platform
ComplyAdvantage
ComplyAdvantageConnector is the right fit when the consequential step is sanctions, PEP, or adverse-media screening at the vendor boundary.
- Use it for: AML name/entity screening and onboarding checks.
- Expected evidence: request/response binding, local pipeline evidence, and surrounding governance context.
- Best when: you need to prove that screening happened, even though the vendor scoring internals remain opaque.
NICE Actimize
NiceActimizeConnector is for AML monitoring, alert triage, and SAR-adjacent review flows where the vendor platform sits in the middle of a regulated operating process.
- Use it for: monitoring, case review, and suspicious-activity workflows.
- Expected evidence: boundary evidence plus optional human-review context around escalations.
- Best when: the platform output feeds a human or downstream workflow you still need to evidence honestly.
FICO Blaze
FicoBlazeConnector is for credit or underwriting logic that runs in a BRMS you can call, but cannot instrument internally.
- Use it for: credit decisioning rules, policy rules, and score-based routing.
- Expected evidence: execution-boundary evidence today, stronger if you later move the governed stage in process.
- Best when: Blaze is the live decisioning layer and you need governance continuity now, not after a platform migration.
IBM ODM
IBMODMConnector covers enterprise decision-management flows where ODM sits between business policy and application execution.
- Use it for: underwriting, eligibility, routing, and other rule-heavy enterprise decisions.
- Expected evidence: boundary execution evidence plus links back to the active control and approval chain.
- Best when: ODM is too central to replace, but still needs explicit governance evidence.
FICO Falcon
FicoFalconConnector fits fraud, thresholding, and alert-generation workflows where the strongest honest claim is often around the decision boundary and follow-on review.
- Use it for: fraud scoring, alert thresholding, and risk escalation.
- Expected evidence: attested or partial execution evidence, often combined with witnessed review steps.
- Best when: you need to show that the fraud-control path ran, not that the proprietary model internals were independently proven.
Pega Customer Decision Hub
PegaDecisioningConnector is for regulated next-best-action or decision arbitration flows where personalization still has compliance consequences.
- Use it for: regulated next-best-action, routing, and customer treatment flows.
- Expected evidence: platform-boundary evidence plus obligation/control linkage around the interaction.
- Best when: you need evidence around how the platform was used without pretending to see inside Pega’s hidden internals.
UpToDate
UpToDateConnector is for clinical decision-support retrieval and supporting evidence in healthcare workflows.
- Use it for: clinical reference lookups and physician-support flows.
- Expected evidence: execution-boundary evidence today, stronger if bounded stages are brought under direct control.
- Best when: you need a clear record of what reference interaction occurred in a consequential healthcare context.
Guidewire ClaimCenter
GuidewireClaimCenterConnector is for P&C claims adjudication and claims operations where the platform remains the system of action.
- Use it for: claims intake, routing, adjudication support, and adjuster workflows.
- Expected evidence: attested boundary evidence plus surrounding review or approval context.
- Best when: you need to prove what was submitted to, or received from, ClaimCenter without claiming direct proof of the hidden platform core.
When A Connector Makes Sense
- the real decision system lives in a vendor platform
- you can call or observe the platform, but not instrument its internals
- you still need evidence that governance-relevant checks ran around that platform interaction
- you need the result to land in the same applicability/obligation/control/approval chain as your direct runtime evidence
What To Expect From Connector Evidence
Connector evidence is usually weaker than direct in-process instrumentation because the vendor logic is a boundary you do not control.
| Surface | Typical Assurance | Why |
|---|---|---|
| Direct application runtime | Proven or Execution | You can instrument the real execution path directly. |
| Vendor platform API boundary | Attested or Witnessed | You can prove the boundary interaction and surrounding context, not the vendor’s hidden internals. |
| Human approval around platform output | Witnessed | The review itself can be bound to a credential and timestamped. |
Connector Families
Financial services
AML, credit, fraud, and regulated decisioning platforms where the governed process crosses a vendor or decision-engine boundary.
Clinical & healthcare
Clinical support and health platforms where integrity, logging, and review context matter even if the platform itself is opaque.
Insurance
Claims, underwriting, rating, and policy administration platforms that still need explicit evidence around consequential decisions.
Other regulated systems
Any consequential platform where you need evidence at the boundary even when you cannot instrument the core runtime.
Connector Integration Pattern
- bind inputs locally
- call or observe the vendor platform
- bind the platform result or event locally
- record the check in your Primust pipeline
- close the run and emit the same governance-linked artifact you would emit for direct runtime evidence
What A Developer Should Decide
For most teams, the connector question is not “which proof backend should I choose?” It is:
- can I instrument this directly?
- if not, what boundary can I prove honestly?
- what human review or manual evidence needs to sit around that platform?
- how does that evidence link back to the approved plan?
Do not oversell connector evidence. If the platform is a black box, say so. Primust is valuable because it makes that limit explicit and still binds what you can prove into the governance chain.
If you can instrument directly, go back to SDKs & Adapters. If you want a path-based starting point, use Common Scenarios. If you want the real package source, use the connectors repo. If you need the exact object and field model connector evidence feeds, use the Technical Reference.