Ai Langchain Nodes

Operator drops an AI Agent node, an OpenAI Chat Model node, and three HTTP Request Tool nodes onto the canvas but only connects the Chat Model. The agent responds without ever calling the tools.

Each tool node must be connected to the AI Agent's `ai_tool` input (the dotted-line port labelled 'Tool'). Verify the agent's exposed tools list reflects all three. Tool nodes provide their `name` + `description` to the LLM — confirm both are set so the model can route.

AI Agent should return `{intent, entities[]}`. Downstream Set node reads `$json.output.intent` but sometimes intent is undefined.

Wire a Structured Output Parser (or Auto-fixing Output Parser) to the agent's `ai_outputParser` input with a JSON schema for `{intent, entities[]}`. Parser validates and (with auto-fixing) re-prompts the model on schema violation. Downstream code should still handle the parsed object's absence defe…

A Pinecone vector store node connected to an AI Agent for RAG returns 3 chunks by default. Some answers cite documents the operator can't find.

Configure `topK` on the retriever explicitly (e.g., 5-10) and pin the vector-store metadata filter for the tenant. Verify that retrieved chunks' metadata (source URL, doc id) is surfaced in the final response so the operator can audit. Use a deterministic embedding model version pinned in the node …