DaraMex Docs

Appearance

AI Module Overview ​

The AI module implements the backend building blocks for configurable AI agents: agent CRUD, chat history, streamed assistant responses, knowledge-base ingestion with RAG, model catalog synchronization, and per-org token usage tracking.

Current Runtime Status ​

  • The module code exists under apps/api/src/modules/ai, but AppModule does not currently import AIModule.
  • AIModule only registers the ai TypeORM connection today. Its controller and provider arrays are still empty, so the controller-defined route surface is not live yet.
  • The docs in this section describe the implemented module behavior in source, plus the current readiness gaps that still prevent end-to-end activation.

Responsibilities ​

  • Manage AI agent definitions, including brain, instruction, knowledge, and limit configurations.
  • Persist chats and messages, then stream assistant responses through the Vercel AI SDK gateway.
  • Run internal subagents for guardrail checks and first-message title generation.
  • Index knowledge documents into Qdrant and expose a RAG tool that agents can call during execution.
  • Mirror the Vercel AI Gateway model catalog into local persistence and track sync status.
  • Record detailed token and cost usage logs, then aggregate monthly organization summaries.

Architecture ​

The module follows the same layered CQRS structure used by the rest of apps/api:

  • Presentation: AgentController, ChatController, MessageController, KnowledgeController, and ModelController define the HTTP surface.
  • Application: command handlers orchestrate agent CRUD, chat lifecycle, streamed execution, knowledge ingestion, and model refresh jobs.
  • Infrastructure: TypeORM persistence, Qdrant, Vercel AI Gateway, embeddings, Redis-backed chat caches, and internal subagents implement the runtime integrations.
  • Domain: Agent, Chat, Message, AgentKnowledgeDocument, Model, TokenUsageLog, and OrgTokenUsageSummary capture the core business state.

Sub-Domains ​