KrisVen sits between enterprise systems and AI applications. It resolves identity, classifies and reviews incoming evidence, applies policy at retrieval time, and records the context that shaped each answer.
The platform is intentionally opinionated about where trust is established. Agents consume governed memory; they do not decide which raw records are safe to reuse.
Documents, APIs, events, and warehouse records enter through tenant-scoped connectors.
People, organizations, accounts, loans, vendors, and relationships are matched to canonical identities.
Sensitive fields, provenance, confidence, and approval state travel with every candidate fact.
Role, tenant, sensitivity, purpose, and point-in-time rules resolve to allow, redact, or deny.
The agent receives citation-ready context while the access and policy decision is recorded.
Connectors normalize source records without making a route or agent responsible for database access. Tenant identity is carried into every ingestion operation.
Memory is more than a vector index. Facts, events, procedures, temporal context, and relationships remain addressable and traceable.
Policy is evaluated before memory is returned. Encryption, sensitivity, retention, and customer isolation are platform controls rather than prompt instructions.
Agents use governed APIs and receive evidence they can cite. Platform teams can inspect retrieval quality, access patterns, and workflow outcomes.
The exact API shape depends on the deployment and domain configuration. The contract below describes the boundary an AI application should be able to rely on.
request = { "tenant_id": "tenant_acme", "actor": {"role": "credit_analyst"}, "subject": "borrower_123", "purpose": "prepare_credit_memo", "as_of": "2026-09-16T14:30:00Z" } response = { "decision": "allow", "facts": [{"value": "...", "source_id": "filing_456", "approval_state": "approved"}], "policy_version": "lending-default-v3", "audit_event_id": "evt_789" }
Implementation boundary: routes and agents should not query storage directly. Repositories own tenant-filtered queries, services own business logic, and the policy engine is called before memory is read.
Bring one agent workflow and let your builders, domain owners, and security team inspect the same path from source record to returned evidence.