Feature Flags
Toggle config-bound features. Changes apply live via the DB override layer.
High risk — real side effects; enable deliberately
GraphRAG code-context enrichment
High
Retrieval
Hybrid Qdrant + Neo4j semantic enrichment injected into codegen prompts. REQUIRES a running Qdrant + Neo4j; a no-op enricher keeps codegen unchanged when off.
GraphRAG:EnabledAGT governance subsystem
High
Governancerequires restart
Trust scoring, ring enforcement, prompt-injection gate, kill-switch, circuit breakers. Changes the agent execution path. Master switch takes effect on the NEXT restart.
Governance:EnabledMulti-tenancy isolation
High
Multi-tenancy
Tenant resolution + per-tenant entity isolation across 41 entities. Flipping this changes query filters — a data-isolation decision, unsafe to toggle on populated single-tenant data.
MultiTenancy:EnabledAutomated dependency remediation
High
Security
Clones repos, bumps vulnerable dependencies, pushes branches and OPENS PULL REQUESTS. Real outward side-effects.
Security:DependencyRemediation:EnabledRegulatory change watcher
High
Regulatory
Background agent that polls registered regulator sources (outbound network), detects changes, and DRAFTS regulatory requirements for compliance review. Makes outbound fetches; drafts are never auto-adopted. Per-source Enabled toggles are separate.
RegulatoryWatch:EnabledMedium risk
Generated-code fidelity check
Medium
Fidelity
Diffs generated code vs the intended design (Missing/Stub findings). Observe = report-only (safe); Enforce (per-pipeline, separate setting) auto-repairs or gates the build.
Fidelity:EnabledDatabase modernization module
Medium
DB modernization
Canonical-IR DB engine (harvest → assess → doc/migrate/logic-to-API). Reads DB catalogs; ExtractLogic (separate setting) triggers logic-to-API codegen.
DatabaseModernization:EnabledLLM Batches API (cost saver)
Medium
Codegen
Routes non-latency-sensitive LLM calls through the Anthropic Message Batches API (~50% cheaper, async latency).
AgenticWorkflow:BatchProcessing:EnabledLow risk
Semantic dead-code analysis
Low
Semantic engine
Reachability-from-entry-points dead-code detection over the real call graph (evidence-only, Approximate).
SemanticAnalysis:DeadCodeEnabledSemantic impact analysis
Low
Semantic engine
"If I change symbol X, here are its callers" impact reports over the call graph (evidence-only).
SemanticAnalysis:ImpactEnabledGraph-ordered migration
Low
Semantic engine
Topological / strongly-connected-component ordering of migration units so each unit has its dependencies.
SemanticAnalysis:MigrationOrderEnabledAuto-resume interrupted runs
Low
Codegen
Crash-safe resume of an interrupted codegen run from its last checkpoint (idempotent).
AgenticWorkflow:Resume:EnableAutoResume