SGR AgentInstall

Summary

Execute an adaptive expert research agent that uses configurable built-in and additional tools to perform iterative research tasks based on user input, optionally using memory and a connected AI language model.

Inputs

• Log Level — Set the verbosity of logging output for the agent's execution, from Debug (most detailed) to None (no logging). Defaults to Error.
• Built-in Tools — Enable or disable specific built-in tools for the agent, including reasoning, planning, clarification, reporting, and finalizing answers.
• Max Iterations — Maximum number of reasoning iterations the agent can perform before forcing completion. Defaults to 10.
• Max Searches — Maximum number of allowed web searches during the research process. Defaults to 4.
• Max Clarifications — Maximum number of clarifications the agent can request. Defaults to 3.
• Prompts — Customize system prompts and templates influencing agent behavior, including system instructions, initial user request format, and clarification response format.
• Additional Tools — Add custom tools with their name, description, and JSON schema to be available alongside built-in tools.
• Options — Additional execution options such as returning the full execution log or returning only the answer text without metadata.
• Chat Model (AI Language Model) (required) — An AI language model node connected to provide the conversational and reasoning capabilities for the agent.
• Tools (AI Tool connections) — Optional connected AI tools usable by the agent during execution.
• Memory — An optional AI memory node to maintain session state across executions, cleared when session changes or on explicit command.

Output shape

A list of JSON objects, one per input item, each containing the agent's answer and optionally detailed metadata or execution log depending on options.

By default, returns a JSON object with answer text and metadata. If 'Answer Only' option is true, returns only the answer text field. If 'Return Full Log' is enabled, includes detailed execution logs. Errors per item are handled according to node failure policy. Multiple input items processed independently.

Examples

Example 1: Research Task

User supplies a research query via input field (e.g. 'task' or 'input'), connects an AI language model as 'Chat Model' input, configures built-in tools and prompts as needed, and optionally provides custom tools and memory node. The agent iteratively plans, searches, clarifies, analyzes, and composes a final answer with citations.

Example 2: Clearing Memory

Set input JSON field 'clearMemory' to true or send 'clear' command in input text to clear connected memory.