AI Building Blocks

Every AI feature in WSO2 Integrator, direct LLM calls, agents, RAG, and natural functions, is composed from the same small set of shared connections and stores. Configure one once, then reuse it wherever it's needed.

Model Providers

Connect to the LLM behind every prompt: the Default WSO2 provider, OpenAI, Azure OpenAI, Anthropic, Google Vertex, Mistral, DeepSeek, Ollama, and OpenRouter.

Embedding Providers

Turn text into semantic vectors used on both ingest and query for similarity search.

Vector Stores

Persist embeddings and run similarity search across In-Memory, Pinecone, pgvector, Weaviate, or Milvus.

Knowledge Bases

The indexable document store RAG reads from and writes to, composed of a vector store, embedding provider, and chunker.

Chunkers

Split documents into chunks before embedding. Smaller chunks improve retrieval precision; larger chunks preserve more surrounding context.

Data Loaders

Read source documents, such as text files or a Microsoft SharePoint library, into ai:Document values ready for chunking and ingestion.

What's next

• RAG — Ground LLM responses in your own documents using these building blocks together
• AI Agents — Use a model provider to power an agent's reasoning
• Direct LLMs — Use a model provider directly, with no agent loop