AI Agent Helicone icon

AI Agent Helicone

AI Agent with integrated LLM provider and Helicone observability

AI Agent Helicone

AI-generated

Summary

Invoke an AI agent powered by a selectable LLM provider (OpenAI, Anthropic, Azure OpenAI, Google Gemini, Groq, Mistral, or OpenAI Compatible) with Helicone observability integration to send prompts, optionally use tools and memory, and receive generated outputs with token usage tracking.

Inputs

  • LLM Provider (required) — Select the language model provider to use (e.g., OpenAI, Anthropic).
  • Model (required) — Specify the model name or deployment to use (e.g., gpt-4o, claude-3-5-sonnet-20241022).
  • Model Options — Configure optional model parameters such as Temperature (randomness), Max Tokens (max generation length), and Top P (nucleus sampling).
  • Prompt Type (required) — Choose whether to define the prompt manually ('Define Below') or take input from a previous node ('Take From Previous Node').
  • System Message — When defining prompt manually, provide the system context message to guide the assistant's behavior.
  • User Message (required) — When defining prompt manually, provide the user input message to send to the AI agent.
  • Helicone Options — Set observability options including Session ID to group requests, User ID to identify the end user, and any custom properties for telemetry.
  • Agent Options — Configure agent-specific behavior such as maximum number of tool-calling iterations and whether to return details of intermediate tool calls.

Output shape

a list of JSON objects, each containing the AI agent's response output; optionally includes intermediate tool call steps and token usage statistics.

Output includes fields 'output' with the generated text or parsed result, 'intermediateSteps' if enabled with details of each tool call and observation, and 'tokenUsage' with prompt, completion, and total token counts. The node supports processing multiple input items, returning one output per item.

Examples

Example 1: Generate AI completions using a specific model with Helicone monitoring

Set provider to 'OpenAI', specify 'Model' (e.g. 'gpt-4o'), define prompt by entering System and User messages, optionally configure model parameters and Helicone options to track sessions or users, then execute to receive AI-generated text and usage metrics.

Discussion