p99 rule-evaluation latency
For pure rule evaluation. The original website presents this as a target—not an end-to-end response-time guarantee.
A decision engine built for AI agents. Execute approved risk strategies with deterministic results, reason codes, and traceable decision records.
Rules · features · model references
Requests · entities · events
Action & score · Reason codes · Trace & audit
PERFORMANCE, IN CONTEXT
A Rust execution core designed for fast decisions. Start with the published figures, then validate your rules, data dependencies, and deployment.
For pure rule evaluation. The original website presents this as a target—not an end-to-end response-time guarantee.
Published in the engine README. Hardware, workload, and test methodology are not specified alongside the figure.
Rust runtime, cached features, and short-circuit rule evaluation.
Distribute traffic across isolated instances. Validate shared-state behavior.
Configure health checks, failover, and recovery. Test before production.
Corint Definition Language
CDL is the common language between risk teams, engineers, Corint Decision, and Corint Work. It covers the full online decision workflow, while staying structured enough for AI-assisted generation and optimization.
CDL defines rules, rulesets, pipelines, registry routing, features, lists, API calls, and final decision mapping in one model.
It can represent login, payment, KYC, credit, AML, review routing, service calls, and feature-backed online decisions.
Unlike lightweight rule DSLs, CDL separates detection, signal generation, orchestration, context, and actions for complete risk workflows.
The declarative structure is readable enough for analysts and structured enough for LLMs to generate, modify, and explain safely.
Corint Work can analyze outcomes, propose CDL changes, optimize thresholds, and prepare strategy updates for review.
Strategies stay in plain YAML, so teams can diff, approve, version, test, and deploy changes through controlled workflows.
Three-layer architecture
Corint Decision separates detection, signal generation, and orchestration so strategies stay reusable, explainable, and easy to operate through code review and version control.
Rules
Atomic risk logic units detect individual patterns and produce scores without deciding final actions.
Rulesets
Rulesets combine rule results, aggregate scores, and produce approve, decline, review, hold, or pass signals.
Pipelines
Pipelines route events, call APIs and services, execute rulesets, and map signals to final results and actions.
Runtime and operations
The project is organized as a Rust workspace that covers parsing, compilation, execution, repository loading, server integration, and API-facing decision execution.
Built as a Rust workspace with core, parser, compiler, runtime, SDK, server, repository, FFI, and LLM modules.
Expose decision execution through HTTP and gRPC server layers with health checks, tracing, CORS, and config loading.
Load CDL assets from a Git-friendly filesystem repository or PostgreSQL-backed repository with versioning patterns.
Reload rules and configuration without rebuilding application code, enabling controlled strategy operations.
Features, lists, APIs
Corint Decision combines request fields with computed features, reusable policy assets, and API responses before pipelines produce final actions and auditable evidence.
Calculate aggregation, expression, lookup, and datasource-backed features before rules run, so every decision has current risk context.
Manage blocklists, allowlists, sanctions data, VIP users, countries, IPs, emails, and other risk assets as governed strategy inputs.
Call internal or external services during pipelines, then return matched rules, scores, reason codes, traces, and processing time for audit.
Integrations
Registry files route login, payment, registration, transaction, or custom events to the right pipeline.
Pipeline steps can call configured external APIs, internal HTTP or gRPC services, and message queues.
Look up online features from Redis or other configured stores and combine them with event context.
Decision results can be persisted for later audit, analysis, debugging, and model or strategy review.