# Iris — Systems Deep Dive

> The `libs/iris/` area: ~262 Nx libraries that make up **Iris**, Oshun's
> AI-assistant domain — the conversation engine, agent runtime, knowledge/RAG,
> four-tier memory, multimodal (voice/vision/BCI/spatial) I/O, emotional
> intelligence, personalization, privacy/safety, platform tooling, and
> multi-language SDKs.

## What this area is

Iris is the platform's conversational and agentic AI substrate, and by project
count it is the largest single lib area in the monorepo. It is not one package
but **262 separate Nx libraries** under `libs/iris/`, each scoped `scope:iris`
(18 tagged `scope:shared` and 1 tagged `scope:oshun`) and almost all implemented
in TypeScript. The packages are organised by capability sub-system rather than
as one monolith, so a consumer can pull in only the slice it needs — e.g. just
`@iris/conversation-core` and one provider, or the full agent + tools +
computer-use stack.

The area layers roughly like this. A **foundation** tier (`@iris/core`,
`@iris/types`, `@iris/config`, `@iris/embeddings`) owns shared types, errors,
context, configuration, and the canonical embedding client. A **conversation**
tier owns multi-turn dialogue (`@iris/conversation-core`),
intent/context/state/style/uncertainty modules, the provider-agnostic
orchestrator (`@iris/conversation-orchestration`), and concrete provider
adapters for Anthropic, OpenAI, Google, and local Ollama models. An **agents**
tier (`libs/iris/agents/`) owns the runtime (`@iris/agents-core`), archetypes,
multi-agent orchestration, a tool framework, and a full
computer-use/desktop-automation stack (including a real multi-crate Rust native
backend). A **knowledge** tier (`libs/iris/knowledge/`) is a production-grade
RAG/GraphRAG/retrieval/grounding stack, and a **memory** tier
(`libs/iris/memory/`) implements a four-tier memory hierarchy plus
personalization and privacy sub-trees. Further tiers cover **multimodal**
(vision, voice, BCI, spatial/XR, IoT), **emotional** intelligence,
**privacy/safety/security**, **accessibility**, **platform** (gateway, billing,
rate-limiting, SDK codegen, white-label), **presence**, **analytics**,
**integrations** with six sibling Oshun domains, **testing**, and client
**SDKs** in five languages.

Honesty note on maturity: the overwhelming majority of these libraries contain
real, domain-specific logic (scoring formulas, retrieval fusion, state machines,
crypto primitives, audio/turn-taking heuristics, statistical tests) and concrete
exported classes/factories — they are not empty scaffolds. Many use **injectable
dependency seams** and in-memory stores so the algorithms run without external
infrastructure, with real backends (Postgres/Qdrant/Redis, vendor LLM/STT/TTS
APIs) wired behind provider interfaces. A small set of nodes are deliberately
**thin V1 product-surface facades** — a single `index.ts` exporting product
metadata, descriptors, and validators (`@iris/accessibility`, `@iris/agents`,
`@iris/conversation`, `@iris/multimodal/vision`, `@iris/multimodal/voice`).
Those are called out as such in their entries below rather than overclaimed.

## How it fits the wider system

These libraries are consumed by the Iris product surfaces (web shell,
mobile/desktop companions, the BFF, and the agent loop) and by sibling Oshun
domains through the `libs/iris/integrations/*` adapters (Hathor, Maya, Nyx,
Psyche, Sophia, Yemaya). The wire contracts for the domain live separately in
`@iris/contracts` (`libs/contracts/iris`), which is the bottom-of-graph schema
package the contracts area documents; the libraries here are the
_implementation_ behind those contracts. Provider adapters compose with the
orchestrator; tools and computer-use compose with the agent runtime;
knowledge/RAG and memory compose into the conversation pipeline;
privacy/safety/accessibility wrap every surface. The five SDKs (`@iris/sdk` and
the Kotlin/Python/Rust/Swift siblings) are the external client view of the same
API. Walk the dependency edges on any node below to see exactly who composes
with it.

## Entity reference

### @iris/a2a

Private, unmounted A2A-shaped compatibility prototype (`libs/iris/a2a/src`):
local agent-card, task-negotiation, and result-sharing experiments only. It is
not a current A2A implementation or interoperability claim. ADR-0091 defers
adoption until a named independent agent boundary exists.

### @iris/accessibility-adaptive

Adaptive-UI accessibility engine (`adaptive-ui-engine.ts`,
`interface-adapter.ts`, `user-capability-detector.ts`, `preference-sync.ts`)
that detects user capabilities and adapts the interface and syncs preferences.
Real implementation, ~8 modules.

### @iris/accessibility-braille

Braille support: `BrailleFormatting`, `BrailleOutput`, `BrailleInput`, and
`braille-display-integration.ts` for refreshable-display I/O and braille
translation.

### @iris/accessibility-cognitive

Cognitive-accessibility services for users with cognitive disabilities — reading
assistance/pace, simplified language, memory aids, distraction reduction,
step-by-step guidance, text highlighting — with named presets (`PRESET_ADHD`,
`PRESET_READING_SUPPORT`, …) and an `applyPreset` composer. Substantial (~14
modules).

### @iris/accessibility-hearing

Hearing-accessibility support for deaf/HoH users: visual alerts, vibration
feedback, real-time captioning, transcript generation, and a sign-language
avatar, composed by `createHearingAccessibilityServices`.

### @iris/accessibility-i18n

Internationalization layer (~16 modules) — language support, RTL layout,
cultural adaptation, localized date/number formats, translation memory,
real-time/UI/content translation — with ~11 locale presets (`I18N_PRESET_AR_SA`,
`_JA`, `_ZH_CN`, …).

### @iris/accessibility-motor

Motor-accessibility / voice-only operation: voice recognition, voice
commands/navigation, dictation, switch access, eye tracking, predictive and
custom input, via `createMotorAccessibilityServices`.

### @iris/accessibility

**Thin V1 facade.** A single `index.ts` exporting accessibility package
metadata, the V2 companion-bridge descriptor (`buildIrisV2AccessibilityBridge`),
capability/mode types, and a profile validator — a product-surface catalog, not
the implementing engines (those are the sibling `accessibility-*` libs).

### @iris/accessibility-testing

Accessibility test tooling: `AutomatedA11yChecks`, `A11yReporter`,
`A11yTestSuite`, and a manual-checklist generator for verifying a11y compliance.

### @iris/accessibility-visual-alternatives

Visual alternatives — `HighContrastModes`, `AudioDescriptions`,
`TextAlternatives`, `SpatialAudioNavigation` — composed by the
`VisualAlternatives` class.

### @iris/accessibility-visual-display

Adjustable-display controls: `TextSizing`, `ColorContrast`, `FontChoices`,
`ReducedMotion`, composed by `createAdjustableDisplay`.

### @iris/accessibility-visual

Visual accessibility / screen-reader support: `ScreenReaderOptimization`,
`ARIALabels`, `FocusManagement`, `KeyboardNavigation`, `AnnouncementSystem`, via
`createVisualAccessibility`.

### @iris/accessibility-voice-ui

Voice-driven UI framework: `VoiceNavigation`, `VoiceFeedback`, `voice-menus.ts`,
composed by `createVoiceUIFramework`.

### @iris/archetypes

Specialized agent archetypes (`libs/iris/agents/archetypes`) —
Research/Code/Data/Creative/ Operations agents (aliased
Researcher/Coder/Analyst/Writer/Operator) over a `base-archetype.ts`, with an
`ArchetypeRegistry` and `createArchetypeWithPersona`. Includes real
`research-nlp.ts`, `web-search-provider.ts`, `command-runner.ts` helpers.

### @iris/computer-use-accessibility

Accessibility-first automation primitives: `SemanticUINavigation`,
`A11yTreeInspector`, `robust-element-selection.ts`,
`accessibility-api-integration.ts` for tree-aware desktop control.

### @iris/desktop-automation

Desktop application automation: `DesktopController`, `WindowManager`,
`MenuNavigator`, `DialogHandler`, `ShortcutExecutor`, wrapped by
`DesktopAutomationAgent`.

### @iris/computer-use-native

**Rust native backend** (`libs/iris/agents/computer-use/native`). A multi-crate
workspace (`iris-desktop-core` traits + per-OS crates for
X11/Wayland/macOS-CGEvent/Windows-SendInput

- an `iris-desktop-napi` napi-rs cdylib). X11 has real input/capture and a Task
  6.4 native fixture; macOS and Windows have input/capture bodies; Wayland and
  every native accessibility-tree walker remain incomplete. The README records
  the per-backend boundary and `#[cfg(target_os)]` selection.

### @oshun/iris-computer-use-native

TypeScript wrapper (`libs/iris/agents/computer-use/native/ts`) around the Rust
desktop binding — `OshunDesktopController` exposes capture, mouse, keyboard,
clipboard, display, and capability methods, while unsupported backend methods
fail closed. `DesktopError` mirrors the native error categories. Task 6.4 proved
the wrapper plus X11 capture/input path on an isolated GTK/Xvfb target; native
X11 clipboard remains unsupported.

### @iris/computer-use

Computer-use GUI-automation agent: `ScreenCapture`, `ActionExecutor`, element
detection, and `createComputerUseAgent`. The directory also nests the Rust
`native/` backend.

### @iris/computer-use-recording

Task recording for computer-use workflows: `ScreenRecorder`, `ActionAnnotator`,
`RecordingPlayback`, `RecordingExport`.

### @iris/computer-use-recovery

Graceful error recovery for automation: `ErrorDetector`, recovery-strategy
selection, `RollbackManager`, `AlternativePathFinder`, orchestrated by
`RecoveryManager`.

### @iris/action-safety

Action-safety system for agents: `ActionValidator`, `DangerousActionBlocker`,
confirmation workflow, undo capability, and audit log, composed as
`SafetySystem`.

### @iris/sandbox

Sandboxed execution environment: `ResourceManager` (with `TokenBucket`),
`NetworkController`, `FilesystemController`, `SnapshotManager`, plus
`SandboxEnvironment`/ `SandboxManager` for isolated agent execution.

### @iris/computer-use-templates

Reusable computer-use task templates: builder, parameterization, sharing, and a
`TemplateLibrary` for lifecycle management.

### @iris/screenshot-vision

Screenshot understanding for agents: `ScreenAnalyzer`, `UIElementDetector`,
`TextExtractor`, `LayoutAnalyzer`, `ChangeDetector`, wrapped by
`ScreenshotVisionAgent`.

### @iris/agents-core

The agent runtime framework (~36 modules):
context/lifecycle/communication/permission managers, `AgentRuntime`,
`AgentSystem`, a `tool-registry`, provider adapters, a cognition gateway,
budget, reward/recommendation models, execution replay/learning, and an
`oshun-ai-bridge`. The substantive heart of the agents tier.

### @iris/agent-marketplace

Marketplace primitives for agent packages: `AgentMarketplace`, `AgentPublisher`,
`AgentInstaller`, `AgentRating` (publish/discover/install/rate).

### @iris/multi-agent-memory

Shared-memory collaboration for multi-agent systems: in-memory
`SharedAgentMemory`, `MemoryBroadcast`, selective-sharing policy, and
memory-conflict resolution.

### @iris/multi-agent

Multi-agent orchestration (~22 modules): `AgentSpawner`, `InterAgentMessaging`,
`TaskDistributor`, `ResultAggregator`, `SharedContext`, plus
consensus/negotiation/ handoff/conflict protocols and an `orchestrator`.

### @iris/agent-personalities

Agent personality profiles: `PersonalityTraits`, `CommunicationStyle`
derivation, and observation-driven `personality-evolution.ts`.

### @iris/ambient

Ambient intelligence (proactive): `BackgroundMonitor`, `AnomalyDetector`,
`OpportunityIdentifier`, `RiskAwareness`, `AlertManager`, composed by
`AmbientIntelligenceSystem`.

### @iris/anticipation

Anticipatory-assistance engine (~30 modules) — pattern recognition, context
triggers, relevance scoring, need/task prediction, productivity-awareness
(break/timing/focus), and opportunity detection — the largest proactive lib.

### @iris/automation

Workflow-automation proposals: `PatternBasedAutomation`, `AutomationSuggester`,
`WorkflowRecorder`, `WorkflowManager`, executed by `AutomationSystem`.

### @iris/reminders

Schedule-aware reminders: `SmartTiming`, `DeadlineTracker`, calendar
integration/sync, and a multi-channel `ReminderEngine` with a `DeliveryProvider`
seam.

### @iris/agents

**Thin V1 facade.** Single `index.ts` exposing the built-in-agent catalog
(`listIrisBuiltInAgents`, `getIrisBuiltInAgent`), capability/intent types,
`selectIrisAgentForIntent`, and a launch-readiness resolver — a product-surface
descriptor layer over the real `agents-*` runtime libs.

### @iris/agent-specialization

Domain specialization for multi-agent systems:
`domain-specialization-training.ts`, `SpecializationTransfer`,
`ExpertiseMapping`, `SpecializationMetrics`.

### @iris/api-tools

API-integration tools for agents: `APIClient`, `RateLimiter`, `OAuthHandler`
(with `OAuthProviders`), `ResponseParser`, composed as `APITools`.

### @iris/tool-auth

Secure tool authentication: `ToolCredentialManager`, `OAuthToolAuth`,
`APIKeyToolAuth`, `CredentialRefresh`.

### @iris/function-calling

Function-calling framework: `ParameterParser`, `ResultHandler`, `ErrorHandler`,
`TimeoutManager`, `FunctionCaller`, composed as `FunctionCallingSystem`.

### @iris/code-tools

Sandboxed code execution for agents: `PythonRunner`, `TypeScriptRunner`,
`ShellRunner` over a `code-executor`/`sandbox` with output capture, composed as
`CodeTools`.

### @iris/tool-composition

Tool-workflow composition: `ToolComposer`, `ToolPipeline`, parallel executor,
and conditional branching for complex tool graphs.

### @iris/database-tools

Database tools for agents: `QueryValidator`, `QueryBuilder`, `DatabaseBrowser`,
`ResultFormatter`, and a `SQLQueryTool` over a `DatabaseExecutor` seam (with a
mock executor for tests).

### @iris/filesystem-tools

File-system tools: `PermissionChecker`,
`FileReadTool`/`FileWriteTool`/`FileSearchTool`, `DirectoryTool`,
`FileWatchTool`, composed as `FileSystemTools`.

### @iris/learning-tools

Tool-learning system: `ToolDocParser` (OpenAPI/TypeScript),
`ParameterInference`, `UsagePatternOptimizer`, dynamic tool registration, plus a
real `TDigest`/`PercentileTracker`.

### @iris/tool-marketplace

Tool marketplace: discovery, installation, rating, and `ToolCertification`
(trust) services.

### @iris/tool-monitoring

Tool-execution observability: `ToolExecutionMonitor`, `ToolErrorTracking`,
`ToolUsageAnalytics`, performance metrics, composed as `ToolMonitoringSuite`.

### @iris/tools-registry

Tool registry: schema management, `ToolValidation`, `ToolDiscovery`, and
versioning, composed as `ToolRegistrySystem`.

### @iris/tool-versioning

Tool version/migration management with a real semver implementation
(`parseSemver`, `compareSemver`, `satisfiesConstraint`), `ToolMigrationUtils`,
deprecation handling, and a compatibility checker.

### @iris/web-tools

Web browsing tools for agents: `HttpClient`, `ContentParser`, and
Search/Fetch/Scrape/ Navigate/Form tools with URL validation, composed as
`WebTools`.

### @iris/workflows

Workflow orchestration (~17 modules): `WorkflowEngine` (with a `TaskExecutor`
seam), `TaskDecomposer`, `DependencyGraph`, `ProgressTracker`, `ErrorRecovery`,
parallel executor, plan visualizer.

### @iris/analytics-ab

A/B-test analytics: `VariantComparison`, `ABTestAnalyzer`, `ABTestReporter`, and
a real `statistical-significance.ts`.

### @iris/analytics-cohort

Cohort analytics: `CohortAnalyzer`, `RetentionAnalysis`, `BehaviorComparison`,
`CohortReporter`.

### @iris/analytics-funnel

Funnel analytics: `FunnelAnalyzer`, `ConversionTracking`, `DropOffAnalysis`,
`FunnelVisualization`.

### @iris/analytics

Privacy-compliant analytics engine (~14 modules): `analytics-engine`,
`event-tracker`, `metrics-aggregator`, retention/usage/quality, plus
launch-monitoring, incident-response, and a `runbook-executor`.

### @iris/analytics-realtime

Real-time analytics: `LiveMetrics`, `StreamingAggregation`, `AlertingTriggers`,
`RealtimeDashboard`.

### @iris/bci-apple-hid

Apple BCI-over-HID integration: `AppleBCIHIDListener`, `IntentSignalDecoder`,
`ThoughtToActionMapper`, `BCICalibrationFlow`, composed as `AppleBCIHIDEngine`.

### @iris/bci-intent-prediction-api

BCI intent prediction: `IntentPredictionModel`, `ConfidenceScorerBCI`,
`AmbiguityResolver`, confirmation flow, feedback loop, and a `fromAppleBCIFrame`
feature-vector adapter.

### @iris/bci-privacy-framework

Neural-data privacy: `NeuralDataMinimization`, `MentalPrivacyProtection`,
`NeuralDataDeletion` (+ in-memory store), `BCIAuditTrail`, cognitive-liberty
consent, and a `BCIPrivacyFramework` with `ProtectedInferenceResult`.

### @iris/code-agentic

Agentic coding: `AgenticCoder`, `MultiFileChanger`, `RefactoringEngine`,
`BugFixer`, `FeatureImplementer` for multi-file code operations.

### @iris/code-architecture

Architecture analysis: `ArchitectureAnalyzer`, `PatternDetector`,
`CouplingAnalyzer`, and `ArchitectureVisualizer` for structural artifacts.

### @iris/code-cli

CLI coding assistant (~25 modules): a `bin.ts` entry, `cli-assistant`, git
workflow (`PRGenerator`, `ReviewAnalyzer`, `commit-generator`,
`conflict-resolver`), build/deploy assistance, log analysis, and `createCLI`.

### @iris/code-codebase

Codebase analysis: `RepoAnalyzer`, `ArchitectureInference`,
`DependencyAnalyzer`, `PatternDetector`, and `CodebaseIndexer`/`CodebaseSearch`,
composed as `CodebaseAnalyzer`.

### @iris/code-consistency

Code-consistency checking: `StyleConsistencyChecker`,
`NamingConventionEnforcer`, `PatternChecker`, `APIConsistencyChecker` with
quick-scan helpers and a combined analyzer.

### @iris/code-dependencies

Dependency intelligence: declaration analysis, `VulnerabilityScanner`,
`UpdateSuggester`, `LicenseChecker` (license-policy validation).

### @iris/code-explanation

Educational code explanation: `CodeExplainer`, step walkthroughs,
`ConceptExtractor`, `DiagramGenerator`.

### @iris/code-generation

Code generation: a TypeScript AST generator with fluent builders
(expression/statement/ function/class), `test-generator`, `doc-generator`, and
contextual autocomplete.

### @iris/code-git

Deep git integration (~21 modules): diff parsing, commit-message/PR generation,
branch-name suggestion, deterministic `conflict-resolver`, style enforcement,
repo indexing, and a technical-debt tracker.

### @iris/code-ide-actions

IDE code actions: `CodeActionProvider`, `FixActions`, `RefactorActions`,
`GenerateActions`.

### @iris/code-ide-completions

Smart completions: context-aware and multi-line completions, completion
explanation, and a `SmartCompletionProvider`.

### @iris/code-ide-hover

Editor hover: `DocumentationHover`, `TypeHover`, `ExplanationHover`, and a
`SmartHoverProvider`.

### @iris/code-ide-inline

In-editor inline chat: `ContextualSuggestions`, `QuickActions`,
`InlineExplanation`, and an `inline-chat-widget`.

### @iris/code-ide

IDE integration layer: a `BaseIDEAdapter` with concrete
VSCode/JetBrains/Neovim/Emacs adapters, JSON-RPC,
inline-suggestions/chat/code-action services, plus `detectIDEEnvironment` and
`createAutoDetectedAdapter`.

### @iris/code-languages

Per-language analyzers extending code-understanding — TypeScript, Python, Java,
Go, Rust, C++, SQL, GraphQL — selected via
`createLanguageAnalyzer`/`hasLanguageAnalyzer`.

### @iris/code-memory-persistence

Coding-style memory: `PreferenceInferrer`, `CodingStyleMemory` (+ in-memory
store), `CodeConventionEnforcer`, project-context persistence, composed as
`CodingMemoryPersistenceEngine`.

### @iris/code-metrics

Code metrics: `CodeMetricsCollector`, `ComplexityScorer`,
`MaintainabilityScorer`, `technical-debt-calculator`.

### @iris/code-quality

Code-quality analysis: `SecurityVulnerabilityDetector`, `PerformanceOptimizer`,
`BestPracticeEnforcer`, `TechnicalDebtIdentifier`, `CodeReviewer`, plus
quick-scan helpers and a combined `QualityAnalyzer`.

### @iris/code-repository-intelligence

Repository intelligence: `RepositoryIndexer`, `DependencyGraphBuilder`,
`ArchitectureInferrer`, `ConventionDetector`, `MultiFileReasoner`, composed as
`RepositoryIntelligenceEngine`.

### @iris/code-review

Automated code-review pipeline: `DiffAnalyzer`, `IssueDetector`,
`SuggestionGenerator`, `ReviewCommentFormatter`, composed as
`CodeReviewAssistant`.

### @iris/code-search

Code search: `SemanticCodeSearch`, symbol search, `UsageSearch`,
`SimilarCodeSearch`, and a `UnifiedCodeSearch` composer.

### @iris/code-semantic

Semantic code analysis: `ControlFlowAnalyzer`, `DataFlowAnalyzer`,
`FunctionPurposeInference`, `VariableUsageTracker`, composed as
`SemanticAnalyzer`.

### @iris/code-turbo-mode

Autonomous terminal execution: `CommandSafetyValidator`,
`AutonomousTerminalExecutor`, `ExecutionResultInterpreter`, an
`IterativeFixLoop`, and a turbo-mode permission manager.

### @iris/code-understanding

Multi-language code intelligence: `AstParser`, `SymbolExtractor`,
`TypeInference`, `CallGraph`, `DataFlow`, composed as `CodeAnalyzer` (the base
the language analyzers extend).

### @iris/concordia-assistant

Iris intake state machine for the Concordia mediation domain (§179.3.1, Phase
179, ~21 source modules): trauma-/coercion-aware safety-signal detection and a
`runSafetyPipeline` routing to Kuanyin restorative circles, hard-boundary gates,
multi-device/party-isolated intake, language detection/drift, and human-reviewer
queues.

### @iris/config

Configuration management: model-provider config (Anthropic/OpenAI/Google),
MemGPT-style hierarchical memory config, voice (STT/TTS/VAD) config, privacy
settings, and a feature-flag system with a `loader`.

### @iris/conversation-citations

Citation/attribution: `CitationTracker`, `SourceAttributor`,
`ReferenceFormatter`, `FactChecker` for inline citations and source tracking.

### @iris/conversation-context

Semantic context tracking: `TopicDetector`, `EntityTracker`,
`ContextCompressor`, `ContextSummarizer`, composed as `ContextTracker`.

### @iris/conversation-core

The core dialogue engine (~28 modules): `ConversationManager` (session
lifecycle), `TurnManager`, `ConversationHistory`, context-window management,
dialogue state, streaming response generator, plus
search-index/bookmarks/highlights/tags/sharing and an in-memory storage backend.

### @iris/conversation-format

Multi-format response formatting: text, markdown, code, structured-data, and
media formatters.

### @iris/conversation-intent

Intent recognition (~36 modules): `IntentClassifier`, `QuestionDetector`,
`UrgencyDetector`, `SlotFiller`, plus an extensive Direct-Command-Control
(`dcc-*`) voice-session suite (intent decomposition, agent-delegation routing,
voice confirmation/handoff, multi-user isolation).

### @iris/conversation-orchestration

Provider-agnostic model orchestration: `ModelOrchestrator`, `RequestRouter`,
`LoadBalancer`, `CircuitBreaker`, `RateLimiter`, and a `ModelRegistry` with
retries/ fallbacks/metrics.

### @iris/conversation-providers-anthropic

Anthropic Claude provider adapter (`anthropic-provider.ts` +
`orchestration-adapter.ts`): chat/streaming/tool-use/vision/extended-thinking
support and a default-config factory, plugging into the orchestrator.

### @iris/conversation-providers-google

Google Gemini provider adapter: chat with system instructions, multimodal
inputs, function calling, streaming, and an orchestration adapter.

### @iris/conversation-providers-local

Local-model provider adapter: an Ollama backend (`ollama-provider.ts`) for
Llama/Mistral/ DeepSeek/Qwen with tool-calling, streaming, and model management.

### @iris/conversation-providers-openai

OpenAI provider adapter: chat/streaming/embeddings/moderation plus an
orchestration adapter and default-config factory.

### @iris/conversation-rag

RAG integration for response generation: `RAGPipeline` with `ContextInjector`,
`SourceTracker`, `RelevanceScorer`, and `RAGFallback` (plus a high-quality
preset).

### @iris/conversation-response

Response-generation pipeline: `ResponseGenerator`, `PromptBuilder`,
`ResponseParser`, `ResponseValidator`, `ResponseFormatter`, and streaming
response support.

### @iris/conversation-state

Advanced dialogue-state management: guard-evaluated state machines,
`CheckpointManager` (persistence/recovery), `BranchManager` (exploration),
composed as `StateManager`.

### @iris/conversation-style

Style adaptation: `ToneDetector`, `FormalityController`, `PersonalityMatcher`,
`StyleLearner`, composed as `StyleAdapter`.

### @iris/conversation-uncertainty

Uncertainty quantification: `ConfidenceEstimator`, `UncertaintyExpressor`
(hedging), `ClarificationGenerator`, `KnowledgeBoundary` detection.

### @iris/conversation-benchmarking

Model benchmarking: latency/quality/cost benchmarks and a `BenchmarkSuite` (with
a mock model invoker) plus a reporter.

### @iris/conversation-branching

Conversation branching: `BranchManager`, `BranchMerger`, `BranchSummarizer`,
`BranchHistory` for alternate dialogue paths.

### @iris/conversation-costs

LLM cost tracking: `CostTracker`, `BudgetManager`, `CostAlerting`,
`CostReporting`.

### @iris/conversation-export

Conversation export: Markdown/HTML/JSON/PDF exporters composed as
`ConversationExporter`.

### @iris/conversation-finetuning

Fine-tuning infrastructure: `DataCollector`, `JobManager`,
`ModelVersionManager`, `finetuning-metrics`.

### @iris/conversation

**Thin V1 facade.** Single `index.ts` for a production-assistant transcript
surface — metadata, `IrisProductionAssistant*` types, and
`create*`/`validate*`/`serialize*` transcript functions — not the conversation
engine (that is `@iris/conversation-core`).

### @iris/conversation-prompts

Prompt management: `PromptLibrary`, `PromptVersioning`, `PromptTesting` (with
`calculateSampleSize`), `PromptOptimizer`.

### @iris/code-reasoning

Code reasoning (`libs/iris/conversation/reasoning/code`): AST/static analysis, a
`RuntimeBehaviorPredictor`, and a `BugHypothesisGenerator`, composed as
`CodeReasoner`.

### @iris/consistency

Self-consistency checking: `ContradictionDetector`, `BeliefTracker`,
`ConsistencyRepair`, composed as `ConsistencyChecker`/`ConsistencySystem` with
quick-check helpers.

### @iris/math-reasoning

Mathematical reasoning: expression parse/eval, symbolic computation,
`EquationSolver`, `UnitConverter` (dimensional analysis), `StatisticalReasoner`,
composed as `MathReasoner`.

### @iris/metacognition

Metacognitive monitoring: `ConfidenceCalibration`, `KnowledgeGapDetector`,
`LearningOpportunityIdentifier`, composed as `MetacognitiveMonitor`.

### @iris/scientific-reasoning

Scientific reasoning: `ScientificReasoner`, `CitationValidator`,
`HypothesisGenerator`, `ExperimentDesigner`, and retraction services.

### @iris/structured-reasoning

Formal structured reasoning: `PremiseTracker`, `LogicValidator`,
`ConclusionDeriver`, `ReasoningExplainer`, composed as `StructuredReasoner`.

### @iris/conversation-search

Conversation search: full-text `ConversationSearch`, semantic/vector search,
filters, and a highlighter.

### @iris/conversation-summarization

Long-thread summarization: `ConversationSummarizer`, `KeyPointExtractor`,
`ActionItemExtractor` (incremental summaries).

### @iris/conversation-templates

Reusable conversation templates: registry, variable substitution, conditional
branching, a fluent builder, and `TemplateExecutor`.

### @iris/conversation-tokens

Token optimization: `TokenCounter`/`SimpleTokenizer`, `TokenOptimizer`,
`ContextCompressor`, `TokenBudgetManager`.

### @iris/core

Foundation library: shared types, error codes/base, context +
correlation/tracer/profiler, a logger, decorators, schemas, and a config loader
(~26 modules) — the bottom of the Iris dependency graph.

### @iris/embeddings

The canonical embedding client for Oshun: a three-tier facade (tier1 OpenAI
`text-embedding-3-large`, tier2 Voyage `voyage-3-large`, tier3 local BGE-M3)
with cache/ disk-cache, metrics, migration, evaluation, an in-process embedder,
and a multilingual benchmark. Decision-doc-grounded
(`docs/releases/p2/embeddings-provider-decision.md`).

### iris/emotional-ethics

Non-negotiable ethical boundaries for emotionally-aware AI: `AIIdentityClarity`,
`CapabilityTransparency`, `CrisisProtocol`, `MandatoryReferral`,
human-relationship respect, composed as `EthicalBoundaries`. (Project name lacks
the `@` prefix.)

### @iris/emotional-multimodal-fusion

Multimodal emotion fusion: `TextEmotionAnalyzer`, `VoiceEmotionAnalyzer`,
`EmotionFusionEngine`, `ContextualEmotionInterpreter`, `EmotionHistoryTracker`,
composed as `MultimodalEmotionAnalyzer`.

### iris/emotional-rapport

Rapport building: `TrustIndicators`, `RelationshipProgress`, personalized
interaction, composed as `RapportBuilder`. (Name lacks `@` prefix.)

### @iris/emotional-recognition

Multi-modal emotion recognition: `EmotionRecognizer`, `TextSentimentAnalyzer`,
`VoiceEmotionAnalyzer`, `MultimodalFusion`.

### iris/emotional-response

Emotionally-adaptive response generation (~16 modules): `EmotionalAdapter`,
`ToneMatcher`, `DeEscalator`, `Celebrator`, `Supporter`, plus
crisis-support/stress-management/mindfulness/ gratitude-journal/wellbeing-report
modules with named default configs. (Name lacks `@`.)

### iris/emotional-social

Social intelligence: `CulturalAwareness`, `FormalityCalibrator`,
`HumorCalibrator`, `BoundaryRespect`, `SocialCueDetector`, composed as
`SocialIntelligence`. (Name lacks `@`.)

### iris/emotional-tracking

Longitudinal emotion tracking: `MoodTracker`, `PatternAnalyzer`, `TrendAnalyzer`
over time. (Name lacks `@` prefix.)

### @iris/emotional-voice-analysis

Prosody-based voice emotion analysis: `ProsodEmotionAnalyzer`, a 48-class
`Emotion48Classifier`, confidence/temporal tracking, micro-emotion detection,
composed as `VoiceEmotionAnalysisEngine`.

### iris/emotional-wellbeing

Consent-based, explicitly non-diagnostic wellbeing monitoring: `GentleCheckIn`,
`ResourceSuggester`, `ConcerningPatternDetector`, `ProfessionalReferral`,
composed as `WellbeingMonitor`. (Name lacks `@` prefix.)

### @iris/ensemble

Multi-model ensemble: `VotingStrategy`, `ConsensusBuilder`,
`DisagreementResolver`, composed as `EnsembleOrchestrator`.

### @iris/failover

Automatic failover for provider integrations: `HealthChecker`, `CircuitBreaker`,
`RetryPolicy`, `FallbackChain`, composed as `FailoverManager`.

### @iris/integrations-hathor

Integration adapter to the Hathor worldbuilding domain: lore context,
`WorldKnowledgeService`, `CharacterDatabaseService`,
`TimelineNavigationService`, `LoreConsistencyChecker`, plus
lore-keeper/plot-advisor/world-consistency agents (~22 modules).

### @iris/integrations-maya

Integration to the Maya metaverse engine: in-world `MayaCompanion`,
`WorldContextManager`, `SpatialVoiceChat`, `WorldCommands`, plus
world-builder/npc-control/asset-creator agents and personality presets.

### @iris/integrations-nyx

Integration to the Nyx astronomy domain: celestial-guide/event/educational
services and stargazer/observation-planner/astronomy-tutor agents, composed by
`createNyxIntegration`.

### @iris/integrations-psyche

Integration to the Psyche domain for AI-avatar presence:
avatar/conferencing/emotion services and presenter/participant/persona agents
with a `platform-adapter`.

### @iris/integrations-sophia

Integration to the Sophia research/knowledge domain: research-assistant,
knowledge-base, literature-search, and citation-manager services + agents,
composed by `createSophiaIntegration`.

### @iris/integrations-yemaya

Integration to the Yemaya creative-studio domain (~27 modules): project-context,
asset-understanding, creative-suggestions, and workflow-automation services +
agents (asset- organizer/audio-mixer/video-editor/design-assistant).

### @iris/knowledge-agentic-rag

Agentic retrieval orchestration: `QueryDecomposer`, `RetrievalStrategySelector`,
`IterativeRefinement`, `SourceQualityAssessor`, dense/sparse/graph retrievers +
fusion + evaluator, plus an `EmbeddingDenseRetriever`, composed as
`RetrievalAgent`.

### @iris/knowledge-chunking

Document chunking for RAG: semantic/hierarchical/code chunkers, table/figure
extractors, and a `ChunkingPipeline`, with Zod-validated config.

### @iris/knowledge

The unified knowledge system (`libs/iris/knowledge/core`, ~18 modules): source
registry, indexing, quality/freshness scoring, conflict detection, citation
formatting, federation, marketplace/monetization, and source-credibility
ranking, with capability probes (`isKnowledgeAvailable`).

### @iris/knowledge-curation

Knowledge curation pipeline: `DuplicateDetector`, `QualityScorer`,
`KnowledgeMerger`, composed as `KnowledgeCurator`.

### @iris/knowledge-embeddings

Embedding-service abstraction with OpenAI/Cohere/local providers, an
`EmbeddingCache`, batch processing, and similarity calculations.

### @iris/knowledge-enterprise

Enterprise knowledge connectors: Confluence/Notion/SharePoint/Google-Drive
connectors over a `BaseConnector` with a `ConnectorFactory` and OAuth2/API-token
auth.

### @iris/knowledge-export

Knowledge export: JSON-LD, RDF, and Wiki exporters composed as
`KnowledgeExporter` (linked- data / semantic-web formats).

### @iris/knowledge-factcheck

Automated fact-checking: `ClaimExtractor`, `EvidenceRetriever`,
`SourceCredibilityScorer`, `VeracityScorer`, composed as `FactChecker`.

### @iris/knowledge-freshness

Freshness tracking: `FreshnessTracker`, `TemporalRelevanceScorer`,
`VersionTracker`, `ChangeNotifier`, continuous index updater, composed as a
`UnifiedFreshnessSystem`.

### @iris/knowledge-graph

Knowledge-graph stack: `EntityExtractor`, `RelationshipExtractor`,
`KnowledgeGraphStore`, `GraphQueryEngine`, `GraphVisualization`, and an
`ingestTextIntoKnowledgeGraph` pipeline.

### @iris/knowledge-graphrag

GraphRAG: `KnowledgeGraphBuilder`, community detection, `GraphSummarizer`
(hierarchical), a `GraphRAGQueryEngine` (multi-hop), composed as
`GraphRAGPipeline`.

### @iris/knowledge-grounding

Grounding/attribution for AI output: claim extraction, source citation,
confidence scoring, fact verification, contradiction + hallucination detection,
composed as `DefaultGroundingPipeline`.

### @iris/knowledge-personal

Personal document indexing: `DocumentIndexer` with file watcher, format handler,
metadata extractor, change tracker, and a privacy filter, integrating with
`@iris/knowledge`.

### @iris/knowledge-query

Query expansion/reformulation for RAG: `QueryExpander`
(synonym/semantic/entity/PRF/LLM), `QueryReformulator`, HyDE
`HypotheticalDocument`, and a `MultiQueryRetriever` with fusion.

### @iris/knowledge-rag-adaptive-chunking

Adaptive chunking: `ContentTypeDetector`, `OptimalChunkSizer`,
`ChunkQualityScorer`, composed as `AdaptiveChunker`.

### @iris/knowledge-rag-debugging

RAG observability: `RetrievalExplainer`, `ChunkInspector`,
`RelevanceVisualizer`, composed as `RAGDebugger`.

### @iris/knowledge-rag-evaluation

RAG evaluation: retrieval/generation/e2e metrics and an `EvaluationDashboard`,
composed as `RAGEvaluator`.

### @iris/knowledge-rag-multimodal

Multi-modal retrieval: `ImageRAG`, `TableRAG`, `CodeRAG`, and a fusion-based
ranker (`MultiModalFusion`).

### @iris/knowledge-rag

Production-grade RAG pipeline: query processing (expansion/decomposition/HyDE),
hybrid retrieval with an in-memory BM25 index + RRF fusion, cross-encoder
reranking, response generation, caching, and evaluation, composed as
`DefaultRAGPipeline`.

### @iris/knowledge-realtime

Real-time knowledge: web/news/academic search and social-media monitoring with a
multi-engine aggregator, composed as a `UnifiedKnowledgeSystem`.

### @iris/knowledge-retrieval

Hybrid retrieval: dense (bi-encoder) + sparse retrievers, multiple fusion
methods, ColBERT-style multi-vector retrieval, cross-encoder reranking, and
evaluation metrics.

### @iris/knowledge-types

Knowledge type system: factual/procedural/conceptual/etc. types, a
`knowledge-factory`, confidence scoring, retrieval strategies, and Zod type
schemas.

### @iris/knowledge-versioning

Knowledge versioning: `ChangeTracker`, `VersionComparison`,
`RollbackCapability`, composed as `KnowledgeVersioner`.

### @iris/mcp

Model Context Protocol client: `MCPClient` with tool discovery, resource access,
and prompt templates for connecting to MCP servers.

### @iris/memory-analytics

Memory analytics: usage stats, growth trends, quality metrics, retrieval-pattern
analysis, and an analytics dashboard.

### @iris/memory-consolidation

Memory consolidation: forgetting-curve `DecayAlgorithm`, `ImportanceScorer`,
`SummaryGenerator`, `MemoryClusterer`, composed as `ConsolidationEngine`.

### @iris/memory-core

The unified four-tier memory system (STM/LTM/episodic/semantic):
`MemoryManager`, `MemoryTierRouter`, `MemoryIndex`, `MemorySerializer`, and
`MemoryGC`.

### @iris/memory-debugging

Memory debugging: `MemoryDebugger`, retrieval explainer, timeline + change
tracking, and a debug-session manager.

### @iris/memory-episodic

Episodic memory: event-based autobiographical episodes with temporal context, a
`MemorableInteractionDetector`, `MilestoneTracker`, task-completion records, and
episodic retrieval.

### @iris/memory-long-term

Long-term memory: persistent `LongTermMemory`, `PreferenceStore`,
`PatternStore`, `CorrectionStore`, plus LTM consolidation/retrieval.

### @iris/memory-migration

Memory import from external AI systems: ChatGPT/Claude/custom importers over a
`BaseMemoryImporter` (with content hashing) and a migration pipeline.

### @iris/memory-persistence

Durable persistence backends: `PostgresStore`, `QdrantStore` (vector),
`RedisCache`, plus backup/restore/recovery (with an S3 backup client), composed
as a `MemoryPersistenceSystem` (constructable from env). The real-infrastructure
layer behind the memory tier.

### @iris/personalization-feedback

Feedback-learning for personalization
(`libs/iris/memory/personalization/feedback`): `FeedbackCollector`,
`ImplicitFeedbackDetector`, `FeedbackProcessor`, `AdaptationEngine`.

### @iris/personalization-inference

Preference inference: `PreferenceInferrer`, `StyleAnalyzer`, `InterestDetector`,
`ConfidenceTracker`, composed as an `InferenceSystem`.

### @iris/personalization

User-modeling and personalization (`libs/iris/memory/personalization`, ~30
modules): preference/interest/goal trackers, `ExpertiseEstimator`,
`BehaviorAnalyzer`, Big-Five detection, context-fusion/prediction,
location/device context, and a `UserModelStore`, composed as
`PersonalizationSystem`.

### @iris/personalization-segments

User segmentation: profile clustering/ranking, segment assignment with
smoothing, segment-default behavior packs, and transition/stability metrics.

### @iris/personalization-testing

Personalization A/B testing: experiment lifecycle, `VariantAssignmentService`,
outcome measurement, and statistical experiment analysis.

### @iris/privacy

Memory privacy/user-control (`libs/iris/memory/privacy`, ~30 modules): memory
viewer/editor/ deleter/exporter/importer, opt-out, consent manager, retention
policy, PII redactor, differential privacy (noise injector, privacy budget), E2E
encryption, key management, and a local store. (Note: distinct from the
`privacy/*` tier libraries.)

### @iris/memory-retrieval

High-performance memory retrieval (<100ms target): hybrid semantic + BM25
keyword search, temporal search, multi-factor relevance ranking, and
multi-source fusion, composed as `MemoryRetriever` (with an embedding-service
seam).

### @iris/memory-semantic

Semantic memory: knowledge graphs, `ConceptUnderstanding`,
`RelationshipTracker`, `DomainExpertiseModel`, `UserKnowledgeGraph`,
`SemanticInference`, composed as `SemanticMemory`.

### @iris/memory-sharing

Memory sharing/collaboration: `MemoryAccessManager`, shared spaces, and
synchronization services.

### @iris/memory-short-term

Short-term/working memory: `ShortTermMemory`, `WorkingMemory`, recent-references
tracking, STM eviction, and temporal decay.

### @iris/memory-tools

LLM-callable memory tools: tool definitions, handlers, and an executor for
letting an assistant manage its own memory (with `formatToolsForClaude`).

### @iris/memory-transitions

Memory-tier transitions: `MemoryPromoter` (STM→LTM), `MemoryDemoter`
(LTM→archival), `ImportanceScorer`, a consolidation scheduler, and transition
metrics.

### @iris/memory-visualization

Memory visualization: graph, timeline, topic-cluster, and relationship view
builders behind a visualization interface.

### @iris/memory-writing

Memory writing: `MemoryWriter`, `MemoryValidator`, `MemoryMerger` (dedup),
`MemoryUpdater`, and a conflict resolver with batch-write support.

### @iris/model-routing

Intelligent model routing/selection: `TaskClassifier`, `CostOptimizer`,
`LatencyOptimizer`, `QualityEstimator`, and configurable `RoutingPolicy`,
composed as `ModelRouter` (with balanced/cost presets).

### @iris/bci

Brain-computer-interface preparation layer (`libs/iris/multimodal/bci`): device
abstraction, neural-signal processing, intent prediction, thought-action
mapping, composed as `BCISystem`, with frequency-band helpers and an
`isBCIAvailable` capability probe.

### @iris/iot

IoT/smart-home integration: device control across
HomeKit/Google-Home/Alexa/SmartThings, room/scene management, automation rules,
and sensor data, with extensive platform/device/ sensor constant tables and
unit-conversion helpers.

### @iris/spatial

AR/XR spatial computing (Vision Pro/Quest/Android XR/HoloLens): XR session
management, spatial anchors, eye/gaze + hand-tracking, plane detection, and a
real 3D-math toolkit (`createPose`, `rayPlaneIntersection`,
`raySphereIntersection`, bounding-box ops).

### @iris/vision-diagrams

Diagram generation from natural language: `FlowchartCreator`, `MindMapCreator`,
architecture-diagram creator, composed as `DiagramGenerator`.

### @iris/vision-documents

Document parsing: PDF/form/receipt/table extraction and layout analysis via a
`DocumentParser`, with a large helper surface (currency formatting, bounding-box
math, reading-order sorting, `tableToCSV`).

### @iris/vision-generation

Image generation/editing: a `generation-impl` with image creation, edit
operations, diagram generation, and screenshot annotation, plus
color/aspect-ratio/file-size helpers and model/ format constant tables.

### @iris/vision-memory

Visual memory: `ImageMemoryStore`, `VisualContextRecall`,
`ImageSimilaritySearch`, `VisualHistoryTimeline`.

### @iris/multimodal/vision

**Thin V1 facade.** Single `index.ts` for a production-footage analysis surface
— metadata, `IrisVision*` types, default required-domains/quality-profile, and
`create*`/`validate*`/ `serialize*` packet functions — not the vision engines
(those are the `vision-*` libs).

### @iris/vision-screenshare

Real-time screen-share pipeline: `ScreenShareReceiver`, `RealTimeAnalysis`,
`ContentUnderstanding`, `ActionSuggestion`.

### @iris/vision-search

Visual search: image indexing, `ImageToTextQuery`, `SimilarImageSearch`,
`ObjectSearch`, composed as `VisualSearchEngine`.

### @iris/vision-understanding

Image understanding: object detection, scene description, OCR (50+ languages),
and chart interpretation via an `ImageAnalyzer`, with IoU/bounding-box math and
OCR-language tables.

### @iris/vision-video

Video understanding: a `VideoAnalyzer` for analysis, action recognition,
temporal reasoning, summarization, and key-frame extraction, with
timestamp/frame-rate/bitrate helpers and codec/resolution tables.

### @iris/voice-auth

Voice biometrics: `VoiceBiometrics`, `VoiceEnrollment`, `VoiceVerification`, and
fallback authentication.

### @iris/voice-commands

Voice command handling: `VoiceCommandParser`, `CustomCommandTraining`,
`CommandConfirmation`, `CommandAliases`.

### @iris/voice-conversation

Full-duplex voice conversation (~12 modules): a `full-duplex-engine` with
VAD/turn-taking/ interruption/backchannel handling, prosody and emotion
sub-modules, and WebRTC voice transport, with rich turn-completion helpers.

### @iris/voice-effects

Voice effects: `VoiceFilter`, `VoiceModulation`, `SpeedAdjustment`,
background-noise addition, with audio utils.

### @iris/voice-journaling

Voice-first journaling: `VoiceJournalRecorder`, `JournalTranscription`,
`JournalSummarization`, `JournalSearch` (semantic/keyword).

### @iris/multimodal/voice

**Thin V1 facade.** Single `index.ts` for a production voice-command surface —
metadata, `IrisVoiceProduction*` types, default command definitions routing to
drone-fleet/switcher/ recorder, and packet `create*`/`validate*`/`serialize*`
functions — not the voice engines.

### @iris/voice-pronunciation

Pronunciation correction: `PronunciationDetector`, custom-pronunciation
dictionary, `PronunciationFeedback`, `AccentAdaptation`.

### @iris/voice

Speech-recognition system (`libs/iris/multimodal/voice/recognition`, ~30
modules): real-time STT (`SpeechRecognizer`), streaming transcription, VAD,
noise processing, accent/dialect handling, multilingual + code-switch detection,
speaker diarization/identification/ verification, anti-spoofing, composed as
`VoiceRecognitionSystem`.

### @iris/voice-synthesis

Text-to-speech (~20 modules): a `SpeechSynthesizer` with voice selection,
prosody/emotion control, and streaming, plus concrete
ElevenLabs/Cartesia/local/persona provider sub-modules, composed as
`SynthesisSystem`.

### @iris/personalization-benchmarking

Personalization benchmarking (`libs/iris/personalization/benchmarking`):
`PersonalizationMetrics`, `ContextLengthStressTest`,
`PersonalizationRegression`, and an `LCMPBenchmarkRunner`.

### @iris/personalization-state-aware

State-aware personalization: `PersonalizedConceptTracker`, `VariationPerceiver`,
a personalization context window, `StateAwareResponseGenerator`, and a
`ContinualPersonalizationLearner`, composed as
`StateAwarePersonalizationEngine`.

### @iris/personalization-user-model-persistence

User-model persistence: snapshot, versioning, migration, export, and
privacy-controls modules over a `user-model-persistence-service`.

### @iris/platform-admin

Enterprise admin-console backend: `EnterpriseAdminService` with a default
config. Small but real (3 modules).

### @iris/platform-billing

Usage billing: `UsageTracker`, `BillingCalculator`, `InvoiceGenerator`, payment
integration.

### @iris/platform-codegen

SDK code generation: `OpenAPIToSDK`, `TypeGenerator`, `ClientGenerator`,
composed as `SDKCodeGenerator`.

### @iris/agent-builder

Custom-agent builder framework (`libs/iris/platform/customization/agents`, ~21
modules): `AgentBuilder`, `ToolBuilder`, `PersonaBuilder`, a `TemplateRegistry`
with built-in templates, plus template marketplace/installer/publisher/rating
and a creation wizard.

### @iris/whitelabel

White-label customization (~24 modules): brand/theme/voice-persona/deployment
builders + managers, a component registry, preset themes/voices, custom-domain
support with a `dns-resolver` and `ssl-provisioner`.

### @iris/platform-gateway

API gateway: `RequestRouter`, `MiddlewarePipeline`, `ResponseTransformer`, and
`APIGateway` with a `FetchUpstreamTransport`.

### @iris/plugins

Plugin architecture (~17 modules): `PluginRegistry`, `PluginLifecycleManager`,
`PluginSandbox`, UI-extension/integration managers, and a plugin workflow
engine, composed as `PluginSystem`.

### @iris/platform-ratelimit

Rate limiting: `RateLimiter`, `TieredRateLimits`, `BurstHandling`, and
`RateLimitHeaders`.

### @iris/platform-reporting

Enterprise reporting: `EnterpriseReportingService` with default metrics-by-type
config. Small but real (3 modules).

### @iris/platform-sdk-docs

SDK documentation generation: `CodeSampleGenerator`, `QuickstartGenerator`,
`TutorialGenerator`, composed as `SDKDocGenerator`.

### @iris/platform-sdk-testing

SDK test tooling: `IntegrationTests`, `CompatibilityTests`, `PerformanceTests`,
composed as `SDKTestSuite`.

### @iris/platform-sdk-versioning

SDK version management: `SemVerManager`, `DeprecationManager`, breaking-change
detector, and a migration-guide generator, composed as `SDKVersionManager`.

### @iris/platform-sla-management

Enterprise SLA management: an `sla-manager` for contract tracking, breach
detection, incident management, and compliance reporting (barrel re-exports
`sla-manager`/`types`).

### @iris/platform-sla

Enterprise SLA management — a smaller sibling of `platform-sla-management`
(barrel re-exporting `sla-manager`/`types`, 3 modules). Honest overlap: both
expose an SLA manager; this one is the leaner variant.

### @iris/streaming

Streaming APIs (`libs/iris/platform/streaming`): SSE, WebSocket, and a
gRPC-streaming abstraction behind a `ProtocolFactory`, composed as
`StreamingManager` with `createStreamingClient`.

### @iris/platform-webhooks

Webhook delivery: `WebhookManager`, `EventDispatcher`, `RetryHandler`,
`WebhookSecurity` (signing/verification).

### @iris/presence-integration

Cross-service presence integration: per-domain integrations
(Maya/Yemaya/Hathor/Nyx + a generic Oshun one) composed as
`UnifiedOshunIntegration`.

### @iris/presence-sync

Cross-device presence sync: `ConversationContinuityManager`,
`TaskHandoffManager`, `NotificationUnificationManager`, `PreferenceSyncManager`,
`StateSyncManager`, composed as `PresenceSyncManager`.

### @iris/privacy-access

Access control (~9 modules): permissions/conditions engine, API-key management,
OAuth2 + OIDC services, audit logging, and a rate limiter — the access-control
layer for privacy-preserving systems.

### @iris/privacy-anonymization

Data anonymization: `KAnonymity`, `LDiversity`, a `DataAnonymizer`, and an
anonymization validator, with real `quasiKey`/`pseudonymize`/`generalizeValue`
utilities.

### @iris/privacy-audit

Privacy audit: `ComplianceChecker`, `AuditReportGenerator`,
`RemediationSuggester`, composed as `PrivacyAuditor`, with risk-scoring
utilities.

### @iris/privacy-communication

Secure communication: `TLSEnforcement`, `CertificatePinning`, `SecureWebSocket`,
and a `SecureGateway`.

### @iris/privacy-dashboard

Privacy dashboard: `DataInventoryView`, `ConsentManagerUI`,
`DeletionRequestsUI`, composed as a `PrivacyDashboardUI`.

### @iris/privacy-encryption

E2E encryption (~14 modules): crypto primitives (AES-256-GCM, ChaCha20-Poly1305,
Ed25519), `KeyManager` (rotation/audit), Double-Ratchet E2E, encrypted
storage/model-IO, secure MPC, and zero-knowledge proofs.

### @iris/privacy-local

On-device AI inference (~23 modules): Ollama and llama.cpp runtimes, a
`ModelManager`/ `ModelOptimizer`, GPU/NPU accelerator detection, offline
cache/sync/features, and hybrid cloud/edge routing, composed as a
`LocalInferenceEngine`.

### @iris/privacy-minimization

Data minimization: `UnnecessaryDataDetector`, `DataMinimizer`,
`RetentionEnforcer`, `AutoPurger`, with end-to-end minimization reports.

### @iris/privacy-private-cloud

Private-cloud compute: `SecureEnclaveProcessor`, `EphemeralProcessing`,
`AuditableCompute`, `HardenedServerConfig` (with a hardened baseline),
`NoDataRetention`, composed as a `PrivateCloudComputeLayer`.

### @iris/privacy-safety-behavior

Behavioral safety/abuse prevention: action confirmation, sandboxed execution,
rate limiting, and anomaly detection, composed as `BehavioralSafety`.

### @iris/privacy-safety-content

Content safety: multi-layer `ContentFilter`, `HarmfulContentDetector`, jailbreak
resistance, age-appropriate responses, `ToxicityDetector`, composed as
`SafetyChecker`.

### @iris/privacy-safety-transparency

AI transparency: identity disclosure, capability-limitation disclosure,
`DecisionExplainer`, uncertainty disclosure, composed as `TransparencySystem`.

### @iris/privacy-safety-validation

Output validation: `PIIDetector`, `BiasDetector`, `FactualityChecker`, composed
as an `OutputValidationSystem`.

### @iris/privacy-secrets

Secret management: `SecretStore`, `CredentialManager`, rotation, and a leak
`scanner`.

### @iris/privacy-security-compliance

Compliance certification for SOC2/HIPAA/GDPR/ISO27001
(`ComplianceCertificationManager`/`createComplianceCertificationManager`). Small
(3 modules) but real.

### @iris/privacy-security-dlp

Data-loss prevention: a single `data-loss-prevention` module re-exported by the
barrel. Thin (the smallest privacy-security lib) but real, not a scaffold.

### @iris/privacy-security-injection

Prompt-injection defense: `PromptInjectionDetector`, `JailbreakDetector`,
`InputSanitizer`, composed as `DefenseStrategy`.

### @iris/privacy-security-logging

Security logging: `SecurityLogger`, `AccessLogger`, `ChangeLogger`,
`SecurityAlertLogger` with default configs.

### @iris/privacy-security-pentest

Penetration testing: `VulnerabilityScanner`, `FuzzTester`, `SecurityReporter`,
composed as `SecurityTestSuite`.

### @iris/privacy-security-threats

Threat detection: `AbusePatternsDetector`, behavioral `AnomalyDetector`,
`ThreatResponse`, composed as `ThreatDetector`, with
`haversineKm`/severity-weight utilities.

### @iris/privacy-sovereignty

Data sovereignty: data-residency controls, GDPR (Arts. 15-22/30/33-34) and
CCPA/CPRA compliance modules, cross-border transfer, and right-to-deletion, with
an extensive typed surface (residency/minimization/consent/analytics) and a
`DataSovereigntyManager`.

### @iris/privacy-third-party-consent

Third-party-model consent: `ThirdPartyModelRegistry`, `ConsentPromptUI`,
`DataMinimizationPreprocessor`, `ThirdPartyAuditLog`, `UserPreferenceStore`,
composed as `ThirdPartyConsentFlow`.

### @iris/privacy-tiered-compute

Tiered compute routing by sensitivity: `ComplexityEstimator`,
`PrivacySensitivityClassifier`, `UserConsentManager`, `TierRouter`,
`TierFallbackChain`, composed as `TieredComputeOrchestrator`.

### @iris/reasoning-thinking

Extended-thinking / chain-of-thought: `ThinkingMode`, `ReasoningChain`,
`ThoughtValidator`, `ThinkingBudget`, and a thinking-summary generator.

### @iris/sdk-kotlin

**Kotlin SDK** (`libs/iris/sdk/kotlin`): a Gradle module with `IrisClient` and
Agent/Conversation/Knowledge/Memory clients, an `HttpClient`, typed `Types.kt`,
and a client test — the JVM/Android client view of the Iris API.

### iris-sdk-python

**Python SDK** (`libs/iris/sdk/python`): an `iris_sdk` package with
`IrisClient`, per-domain clients (agent/conversation/knowledge/memory),
`http.py`, typed `types.py`, and `py.typed`, packaged via `pyproject.toml`.
(Project name lacks the `@iris/` prefix.)

### @iris/sdk-rust

**Rust SDK** (`libs/iris/sdk/rust`): a `iris_sdk` crate (`lib.rs`) with an async
`IrisClient`, per-domain client modules (agent/conversation/knowledge/memory),
an HTTP layer, typed errors/types — type-safe async access to the API.

### @iris/sdk-swift

**Swift SDK** (`libs/iris/sdk/swift`): a SwiftPM package (`IrisSDK`) with
`IrisClient`, Agent/Conversation/Knowledge/Memory clients, an `HTTPClient`,
typed `Types.swift`, and tests — the Apple-platform client.

### @iris/sdk

The official TypeScript/JavaScript SDK (`libs/iris/sdk/typescript`): an
`IrisClient`/`Iris` client with `createClientFromEnv`,
conversation/agent/memory/knowledge sub-clients, and an HTTP layer
(`client.chat(...)`).

### @iris/testing-chaos

Chaos testing: `NetworkChaos`, `ServiceChaos`, `DataChaos`, composed in a
chaos-test framework.

### @iris/testing-load

Load testing: `ScenarioBuilder`, `LoadProfiler`, `LoadReporter`, composed as a
load-test suite.

### @iris/testing

The Iris test-utilities library (~30 modules): mock model/memory providers,
fixtures, conversation/user factories, assertions, benchmarks, and
API/event-contract + bias/ factuality/consistency/latency/load test helpers —
the shared testing toolkit for the area.

### @iris/testing-synthetic

Synthetic monitoring: `SyntheticMonitor`, health-check probes, user-journey
tests, and alert integration.

### @iris/testing-visual

Visual regression testing: `PixelDiffAnalyzer`, screenshot comparison, a
visual-test framework, and a regression reporter.

### @iris/types

Shared Iris type definitions: conversation/memory/agent/model/user/tool/event
types — a contracts-layer (`layer:contracts`) package consumed across the area.

### @iris/voice-empathic-synthesis

Empathic voice synthesis (`libs/iris/voice/empathic-synthesis`):
`EmotionalToneAdapter`, `ProsodyModulator`, `EmpatheticMirroring`,
`DynamicToneShift`, `FrustrationDeescalation`, composed as
`EmpathicSynthesisEngine`.

### @iris/voice-providers

Voice provider/profile layer (`libs/iris/voice`): a provider registry, a
`profile-registry`, voice cloning workflow, and integrity tooling
(`watermark.ts`, abuse detection) — barrel re-exports
`providers`/`cloning`/`integrity`.

### @iris/voice-ultra-low-latency

Ultra-low-latency voice (~18 modules): a `StreamingTTSPipeline`,
`AudioBufferPreloader`, `LatencyMonitor`, full-duplex interruption handling,
endpointing/turn-taking/backchannel models, and a
`VoiceActivityOptimizationEngine`, composed as `UltraLowLatencyVoiceEngine`.