How to Use the AI Agent Action in Sparkforge Workflows

The AI Agent action runs inside Sparkforge Workflows. You give it instructions and a tool set, and it decides what to do, in what order, and with what inputs at execution time. One AI Agent action can replace a long chain of individually configured steps.

Note: AI Agent is a premium action. Each execution incurs additional charges.

Why use it

• Autonomous execution: the agent picks tools and inputs based on your instructions.
• Faster building: one action replaces many manually configured steps for the right use cases.
• Flexible setup: start from a template or build your own.
• Downstream output: return text or JSON for use in later steps.
• Visibility: Execution Logs show reasoning, tool calls, and token usage.

Configuration fields

• Action Name: custom label for the action.
• Template: pre-built configuration that fills Instructions and Tools, or pick Build Your Own.
• Instructions: what the agent does, how it reasons, the rules it follows. The most important field.
• Enhance Prompt: restructures loose instructions into a more organized prompt. Reversible.
• Model: which AI model powers the agent.
• Tools: the actions the agent is allowed to execute.
• Conversation Memory: rolling summary of past runs for the same contact at this step.
• Output format: None, Text, or JSON.

Templates

Templates pre-fill instructions and tools for common scenarios. Templates that need a specific trigger appear unavailable until that trigger is added to the workflow. Available templates:

• Build Your Own
• Form Lead Follow Up
• No Show Appointment Recovery
• Facebook Lead Nurturing
• Stale Deal Nudge Agent
• Lead Research & Enrichment
• New Appointment: Enrich & Confirm
• Quiz Lead Scoring & Routing
• Call Transcript Summary & Action Items
• Task Creation from Won Deals
• Instagram / Facebook Comment-to-DM
• Lead Pipeline Tracker

Writing instructions

Strong instructions usually include:

• A role statement
• Context about the trigger data the agent will receive
• A step-by-step workflow
• If/then decision logic
• Messaging guidelines
• Constraints (what the agent should not do)

If you are unsure how to structure the prompt, write it in plain language and click Enhance Prompt. The editor supports rich text formatting including headings, bold, lists, code blocks, and block quotes.

Configuring tools

Click + Add Tool to open the picker. Tools are organized by category and searchable. Each tool you add appears in the configuration panel and can be set up individually.

Field-population behavior:

• Let AI decide all field values: ON — the agent generates values at runtime from instructions and workflow data.
• Let AI decide all field values: OFF — set some values manually; the agent fills any blank ones.

Each field also has an individual toggle for AI vs. manual.

How to set up the action

1. Go to Automations > Workflows.
2. Create or edit a workflow.
3. Add a relevant trigger (Form Submitted, Contact Created, etc.).
4. Click + Add or select Please select action.
5. Pick AI Agent from the Workflow AI section.
6. Enter an action name.
7. Pick Build Your Own or a template.
8. Add your instructions. Click Enhance Prompt if needed.
9. Choose a model.
10. Click + Add Tool and configure each tool.
11. Open Advanced Options for Conversation Memory and Output Format.
12. Save the action.

Conversation Memory

Off by default. When on, the agent uses a rolling summary of past runs for the same contact at this step. Useful for recurring or multi-touch workflows like periodic lead follow-up. For one-off workflows like a single form follow-up, leave it off.

Output Format

• None: the agent acts but returns no structured output.
• Text: free-text response. An Output Description field lets you describe what should be in the response.
• JSON: structured data based on a schema you define with named properties, types, and optional descriptions. Use this when later steps need to reference specific values.

Execution Logs

After a run, open the Execution Logs tab in the workflow builder and select an AI Agent execution. The trace can include:

• Agent Input: the full input the agent received.
• LLM: each model call used to reason about the next step.
• Tool executions: each tool the agent called, including duration and success or failure.

Expand any step to inspect input and output. The log view supports both table and raw JSON formats.

Pricing

Total cost per execution comes from two parts:

• LLM token usage: the agent may make multiple model calls per execution. Token totals depend on instruction length, task complexity, and tool calls.
• Premium action executions: if the agent calls premium actions as tools, those run at the standard per-execution rate for those actions.

Standard built-in tools like Send SMS, Update Contact Field, and Add Tag do not add a premium charge beyond normal platform usage. Token usage is visible in Execution Logs; billing details are in the account billing section.