AI Agent PhoenixInstall

Summary

Invoke an AI agent using a selectable LLM provider and model with customizable prompt, agent behavior, and Arize Phoenix observability integration to trace and monitor interactions.

Inputs

• provider (required) — Select the LLM provider to use from available options like OpenAI, Anthropic, Azure OpenAI, Google Gemini, AWS Bedrock, Groq, Mistral, Ollama, xAI Grok, vLLM, or OpenAI Compatible.
• model (required) — Specify the model or deployment to use for the chosen LLM provider. The allowed values and defaults depend on the provider.
• modelOptions — Configure model parameters such as Max Tokens (max tokens to generate) and Temperature (controls output randomness).
• promptType (required) — Choose how to provide the prompt, either Define Below (enter messages manually) or Take From Previous Node (use data from prior node).
• systemMessage — Define a system message setting context or behavior for the AI assistant, used if promptType is Define Below.
• text — Input user message or prompt text, used if promptType is Define Below; can reference previous node data dynamically.
• phoenixOptions — Options to group and attribute traces for observability, including Session ID, User ID, and filtering Tags.
• agentOptions — Agent control parameters such as Max Iterations (limit agent iterative steps) and Return Intermediate Steps (whether to output agent's intermediate reasoning steps).

Output shape

a list of JSON objects, each containing the agent's output under 'output', optionally 'intermediateSteps' if agent iteration steps are returned, and token usage info under 'tokenUsage' if available

The output is paired with the input item index. The node handles multiple input items. If an output parser is connected, the raw LLM response is post-processed. The node integrates OpenTelemetry tracing spans recorded to Arize Phoenix for observability, capturing LLM inputs, outputs, and tool calls. Errors per item are emitted in the corresponding output object if Continue On Fail is enabled.