AI Agent + LangfuseInstall

Summary

Run an AI agent with Langfuse tracing using a connected chat model (and optional fallback), tools, and memory. The system prompt is sourced either from a Langfuse prompt (with variable substitution) or defined manually. Input text comes from the previous node's chatInput field (Auto) or the Text parameter (Define). The node streams results if enabled, processes items in batches, can return intermediate steps, and passes through binary images.

Inputs

• main (required) — Data items to process; each item may contain a chatInput field used when Prompt Type is Auto.
• ai_languageModel (required) — Chat Model – the primary language model that supports tool calling.
• ai_languageModel — Fallback Model – used only when Enable Fallback Model is active.
• ai_memory — Memory – optional chat history persisted across agent runs.
• ai_tool — Tool – one or more tools available to the agent.
• ai_outputParser — Output Parser – optional parser to enforce a specific output format (requires Require Specific Output Format).

Output shape

main output containing the agent's final response; if Return Intermediate Steps is enabled, includes an intermediateSteps array.

When Prompt Source is Langfuse, the selected prompt's system message (with variables substituted) overrides the System Message option; if the prompt includes a user message, it replaces the Text/chatInput regardless of Prompt Type. Binary images from input data are appended as image_url content when Passthrough Binary Images is true. The node respects batch size and delay between batches, and streams token‑by‑token when Enable Streaming is true and the execution environment supports it.

Examples

Example 1: Simple Q&A with Langfuse‑managed prompt

Prompt Source = Langfuse, select a prompt, leave Model Source = Langfuse, leave Prompt Type = Auto, connect a chat model, optionally add memory and tools, run.

Links

• n8n community nodes installation guide