Browse the docs

Imported from laraai/docs/GLOSSARY.mdEdit it in the module repository to change this page

AI module glossary

Canonical English names for AI entities in this module. Use these terms in code, APIs, and cross-module documentation.

Chat and conversations

Term Meaning
Conversation Persistent chat session owned by a user; holds ordered Message rows.
Message Single user or assistant utterance within a Conversation.
ConversationSummary Condensed history used to keep context windows manageable.
ChatService Creates conversations and builds the protected agent (buildProtectedAgent). It does not answer: every message endpoint calls InAppAssistanceService.
ChatController HTTP layer for the conversation endpoints (insertMessage, streamMessage, listing and deleting).
streamMessage Answers 422 in_app_streaming_unavailable: the protected assistant validates the whole answer before it sends any of it. The event stream is POST /app/ai/agent (AgentController, AG-UI events), which sends the message once, whole.
insertMessage Non-streaming JSON response path (jobs, integrations, tests).
ChatAgent NeuronAI agent wrapper that every call to the model goes through (forFeature(), ask(), structured()); the protected agent of ChatService is one.

Tool system (governed writes)

Term Definition
Write proposal (WriteProposal) A write the assistant asked for, stored with its exact payload; it changes nothing until the person confirms it.
WriteProposalService Stores proposals, applies one when the person confirms it (once, under a row lock) and logs each step.
AssistantWriteController The person's confirm, reject and show actions on a proposal, the only way a proposal is applied.
governed_writes / crud_reads Policy capabilities that admit the crud_* write and read tools for the in-app profile only.
unmoderated_writes Operator list of entities without approvals on which the assistant may nonetheless write; "applied directly, no vote".
ToolResultGuard Withholds from the model the result of an entity read tool whose text reads like an instruction.
ActionRequest, RiskClassifier Removed 2026-10-07. A tool-level approval path that nothing used; approval is Core's, at the model.

RAG and documentation intelligence

Term Meaning
DocumentationAgent NeuronAI RAG subclass: chunking, embedding, retrieval, answer synthesis.
DocumentationService Application service for ingest, reindex, and query over module docs.
EmbeddingService Implements IEmbeddingService; produces vectors for models and doc chunks.
rag_paths() Helper returning documentation roots and source prefixes for indexing.
FileVectorStore Filesystem-backed vector store for RAG persistence.
MemoryVectorStore In-memory vector store (tests / ephemeral).
MarkdownAwareSplitter Default SplitterInterface; keeps fenced blocks (e.g. Mermaid) intact.
SplitterFactory Resolves the active document splitter implementation.
reindexBySource Incremental RAG update per logical source prefix.
indexDocuments Artisan-driven full or incremental documentation indexing pipeline.

Core search orchestration (AI bindings)

Term Meaning
HandleModelIndexingListener Listens to ModelRequiresIndexing; dispatches GenerateEmbeddingsJob when enabled.
GenerateEmbeddingsJob Embeds searchable model content; emits ModelPreProcessingCompleted('embeddings').
HandleModelTranslationListener Listens to TranslatedModelSaved; dispatches TranslateModelJob.
TranslateModelJob Auto-translates translatable models when configured.
CrossEncoderService Optional IReranker binding for search result reranking.
SearchOrchestratorAgent Optional ISearchPlanner binding for multi-step search plans.
LlmQueryIntentParser Optional IQueryIntentParser binding.
SearchEmbedder Optional ITextEmbedder binding for query/document embeddings in Core search.

Moderation (approval workflow)

Term Meaning
ModerationService LLM analysis of pending Modification records; returns structured verdict.
ApproveModificationJob Applies AI vote on a Modification as the configured system user.
HandleModificationModerationListener Entry listener on ModificationRequiresModeration.
ModerationContextBuilderRegistry Core registry; AI resolves context without importing domain modules.
ModerationResult Structured approve / reject / uncertain outcome from ModerationService.
ModerationApprovalMode Policy enum: threshold, dual, uncertain-fallback variants.
ai.features.moderation.entities.{table} Per-entity AI moderation switch, declared by the AI module for each model with a registered moderation adapter.

Contextual suggestions

Term Meaning
ContextualSuggestion Stored suggestion tied to UI context for proactive assistant hints.
SuggestionController HTTP layer for contextual suggestion endpoints.

LLM providers

Term Meaning
EmbeddingsProviderFactory Resolves embedding provider from config('ai.*').
Neuron AI The agent framework (neuron-core/neuron-ai v4) behind every model call: providers, tools, RAG, structured output. It replaced LLPhant, which is no longer a dependency.
  • docs/ARCHITECTURE.md — module layout and message flows
  • docs/DESIGN_DECISIONS.md — why the answer is sent whole, governed writes, memory, internal use
  • docs/SEARCH_AND_TRANSLATION.md — indexing and translation listeners
  • docs/MODERATION.md — moderation pipeline and approval modes
  • Modules/Core/docs/EVENT_ORCHESTRATION.md — Core event bus contracts
100%