Capability Guides

These guides cover the major subsystems that give Beluga AI agents their capabilities. Each guide explains the architecture of a subsystem, the design patterns it uses, and how to integrate it into your applications. They assume familiarity with the Foundation guides and build on the core patterns established there — registries, middleware, hooks, and streaming.

Every capability follows the same extensibility model: a small Go interface defines the contract, providers register via init(), and middleware wraps behavior without modifying implementations. This consistency means that once you learn one subsystem, the patterns transfer directly to the others.

Guide	Description
RAG Pipeline	Build retrieval-augmented generation pipelines with embeddings, vector stores, and advanced retrieval strategies like hybrid search, CRAG, and HyDE
Document Processing	Load, parse, and chunk documents from multiple sources and formats for RAG pipelines and knowledge bases
Memory System	Implement persistent agent memory using the MemGPT-inspired 3-tier model — Core (always in context), Recall (conversation history), and Archival (vector-searchable long-term storage)
Tools & MCP	Create typed Go tools, organize them in registries, and connect to remote MCP servers for runtime tool discovery and interoperability
Voice AI	Build real-time voice applications using a frame-based processing pipeline with STT, TTS, S2S, VAD, and pluggable transport layers
Multimodal	Process images, audio, and video with multimodal language models for document intelligence, visual Q&A, and content analysis