Superagento.
Superagento

Build your first AI agent with a result you can check.

To build an AI agent in Superagento, add an AI Agent node, connect a Chat Model and Agent Tools, and define the task. Start with a calculation you can verify. Once the agent returns the expected structured result, connect that output to the next step in your workflow.

Superagento guide · Updated

Use this guide alongside the current product documentation. Available models, integrations and allowances depend on your configured account.

What do you need before you start?

Create a Superagento account and open a new workflow. Have access to a configured model through your plan or a supported personal API key. The AI Agent needs live provider access; it does not manufacture a simulated response when credentials are missing.

For this first example, use a manual trigger and the calculator tool. It avoids writing to an external service and gives you an answer you can check independently: 125 multiplied by 4, plus 80, equals 580.

Connect the model, tools and output

Add AI Agent to the canvas. Connect Chat Model to its model port and Agent Tools to its tools port. Enable Calculator. Connect Structured Output and define a required numeric field named total. Leave Session Memory disconnected because each calculation is independent.

Task: Calculate 125 * 4 + 80 using the calculator.
Return a JSON object with a numeric total.
Expected result: {"total": 580}
  1. Give the agent a narrow role: calculate the requested expression using its calculator.
  2. Choose an available model and credential source in Chat Model.
  3. Enable Calculator in Agent Tools.
  4. Set Structured Output to a record with total as a required number.
  5. Save the workflow and run a test.

Inspect the result, not just the success badge

Open the run details and inspect the agent's tool call and returned JSON. Confirm that the total is numeric and equals 580. If the result is text instead of the expected record, review the output configuration and instructions before wiring it into a later action.

Try another expression and an incomplete request. Decide what a valid failure looks like. For an incomplete business input, asking for clarification or marking a record for review can be more useful than attempting to infer information that was never supplied.

Map the answer into a workflow action

A downstream step can read the structured result using the agent's node ID. Replace AGENT_ID in the example with the ID from your canvas. Field mapping makes the data dependency explicit, so changing a prompt does not mean copying values by hand between steps.

{{steps.AGENT_ID.json.total}}
  1. Add a destination action only after the standalone result passes your checks.
  2. Map the total field to the destination's matching field.
  3. Use a test spreadsheet or another reversible destination for the first write.
  4. Review the entire execution before moving to a schedule or webhook.

Expand the task one capability at a time

Add reference text or an approved HTTP GET endpoint only when the task needs more information. Add memory only when separate requests should share conversation context. A trusted conversation ID is essential: using one ID for unrelated people mixes their context.

Retest the original examples after each change. Keep the tool set small and avoid putting credentials into instructions. Superagento stores supported personal provider keys separately from the workflow graph; a shared workflow is not a reason to copy a key into a prompt.

Frequently asked questions

Why does my agent need an Agent Tools node?

Superagento requires an enabled tool set for AI Agent. For a simple transformation without tool use, choose an ordinary AI step instead.

Why did my first agent test stop?

Check its model connection, provider access, enabled tools and required input. Run details identify the failed stage; do not assume that missing credentials will produce a demo answer.

Should I connect memory for every agent?

No. Independent requests usually do not need memory. Connect it when previous messages are part of the task, and scope the conversation ID to the intended conversation.

Can the agent return a list?

Yes. Choose a list in Structured Output and connect a Loop when each returned item needs its own action. Check list fields before processing a live batch.

Put the idea into practice

Start with a small test and a result you can verify. Keep the first destination reversible, then inspect the complete run before turning on automatic actions.

Build with Superagento Follow the quickstart

Related guides

Product reference