Overview

The Agent step uses natural language processing to analyze content, extract information, and generate structured outputs that can be used in downstream steps.

Use When

• You need respond to a question based on analysis of its content

• You want to incorporate references to existing sources within the agent's capabilities

• You want to extract specific data from contracts or documents

• You need to analyze content and generate recommendations

• You want to classify or categorize information

• You need suggestions that can be reviewed and refined

Configuration Options

Context

Specify any accessible resources or reference materials the Agent should consider during analysis. All data stores in Chamelio Core could be leveraged by the agent.

Instructions

Provide natural language prompts that describe what you want the Agent to do. Be specific about what information to extract or what analysis to perform.

Outputs

Define the output fields that will be available as @AgentStepName.OutputFieldName for use in downstream steps. Outputs must be explicitly defined so they're available as variables throughout your flow.

Optional Human Approval

Configure the Agent step to route its suggestions to a human approver before passing results downstream. This ensures quality control over AI-generated outputs.

Example

Configuration

Instructions: "Extract the contract term, renewal date, and any non-standard payment terms."

Outputs defined:

• ContractTerm

• RenewalDate

• PaymentTerms

Result

When this Agent step runs, its outputs become available as:

• @AgentStepName.ContractTerm

• - @AgentStepName.RenewalDate

• - @AgentStepName.PaymentTerms

These can be used in subsequent steps for approval, document generation, notification, or conditional branching.

Notes

• Outputs must be explicitly defined during configuration

• Outputs are suggestions to be reviewed, not final decisions

• Agent steps can include optional human approval checkpoints

• Agent outputs become variables available throughout the remainder of the flow

• Consider the quality and clarity of your instructions—more specific instructions typically yield more accurate results