Index
AI terms, capabilities, and workflow language.
Every term we use, defined plainly. Substrate, runtime, evals, governance, retrieval, workflow patterns, and closed-loop operations — for non-experts, start with Information AI, RAG, MCP, and Evals.
A
A2AAdaptive groupingAdaptive reasoningAdaptive visualization selectionAgent CardAgent HarnessAgent MemoryAgent ObservabilityAgent Protocol StackAgent RegistryAgent RuntimeAgentic WorkflowAgents on a rosterAI GatewayAi Ready StorageAI-driven layout intelligenceAI-Native DashboardsAI-native widget architectureAnalytics WorkloadsApproval GateApproval InboxAutonomy with boundaries
C
Capability mapChannels in scopeChannels we typically supportChat Orchestration RuntimeCitation QualityClosed loop with the workClosed-Loop KnowledgeClosed-loop listeningCommon stacksCommon substratesCompactionComposition of a useful eval setConclusionConnection to forensics and the learning loopConsent and retentionConsent and the off-switchContext BudgetContext EngineeringContext engineering as the differentiatorContext overviewContext-aware linkingConversation ForensicsConversation IntelligenceConversation ListenersConversational interfacesCore philosophyCorrection CaptureCost overviewCost Reduction
H
How extraction is runHow it actually worksHow it actually works — the mentor–apprentice loopHow it differs from a knowledge graph or lineage graphHow it runs on AWSHow optimization worksHow they are builtHow we clusterHow we enforce itHuman ApprovalHuman In The LoopHuman Review QueueHybrid Retrieval
R
RAGReflection and validationRegression DatasetsReplaying the thread around the momentRerankingReranking PolicyResearch notesResolution QaResponses APIRetrieval is a system, not a database callRetrieval ReadinessRetrieval-augmented generationRisk ReviewRoot Cause AnalysisRouting axesRuntime parameters
S
Safe DeploysSavings suggestionsSelf-Optimizing AgentsSemantic navigationSentiment TrendsSignal ExtractionSkill DistillationSlack To KnowledgeSource ContractsSource GraphSource Of TruthSources we build fromSpatial dashboard navigationStacks we useStandards we map toSub Agent ArchitectureSubstrates we build onSupport Thread AnalysisSupport TriageSynthetic Personality
T
Task LifecycleThe blank canvas paradigmThe cognitive runtime layerThe cognitive widget runtimeThe prompt-native widget conceptThe substrate determines the agentThe visibility imperativeTightening over timeTool Audit TrailsTool CallingTool ExecutionTool PermissionsTool Result ClearingTool Schema ContractsToolingTools and MCP executionTools connectedTools that fitTrace everythingTrace exposureTrace GradingTrace Replay
W
Webchat SignalWebSocket ModeWhat 'LLM-ready' actually meansWhat a contract specifiesWhat a diff measuresWhat a runtime providesWhat can changeWhat downstream uses it forWhat forensics produces as a learning artifactWhat gates encodeWhat gets evaluatedWhat is enforcedWhat it isWhat it is — and what MCP isWhat it is notWhat it works withWhat makes a skill document hold upWhat we are buildingWhat we buildWhat we evaluateWhat we extractWhen the runtime mattersWhen you need itWhere the loop closesWhy a registry, not just a tool listWhy contracts beat best-effortWhy diffs, not single scoresWhy it mattersWhy it matters for AIWhy it needs the eval setWhy routeWhy this changes dashboardsWhy this paper existsWhy traces feed evalsWidget generation through languageWorkflow EvalsWorkflow MutationWorkflow Runtime