Skip to content
Open Agents Builder

Building Workflow based agent - describing products from pictures

🛠️ What you’ll build

A Workflow (API) agent that:

  1. accepts a product photo from the user,
  2. lets a sub-agent describe the item in natural language, and
  3. returns a ready-to-run Shopify GraphQL productCreate curl command you can paste into a terminal (or call programmatically) to add the product to your store.

The whole process takes 5–10 minutes and zero custom code.

1. Create the Workflow agent

StepActionWhere
1Log in to openagentsbuilder.com (a free trial is enough).–
2Press New Agent… in the header.Top bar
3Give it a name, e.g. “Import product from the picture”.General → Agent Name
4Change Agent type to App / Workflow [API].General section
5Click Save.Bottom left

Why a Workflow agent?
Chat agents expose a conversational endpoint; Workflow agents expose flows that you can:
• run in a lightweight UI you can share with anyone, or
• call from any backend over HTTPS.

###. 2 Add a sub-agent that “understands” images

  1. Go to Sub-agents ➜ Add agent….
  2. Name it “Photo descriptor”.
  3. Leave the model as gpt-4o (or whichever model you prefer).
  4. Paste a short System prompt – e.g.:
You're great at extracting the features of a product from a picture
and generating structured descriptions.
  1. Save.

No extra tools are needed here because the only job is to convert an image into text + a GraphQL snippet. If you later want the agent to hit the Shopify API directly, you could add an “HTTP request” tool.

3. Design the flow

3.1. Create the flow shell

  1. Open Flows ➜ Add flow.
  2. Name it Describe product – set the Flow code to describe-product (this string becomes part of the API URL).
  3. Save to reveal the flow designer.

3.2. Define inputs

  1. In the Inputs accordion press Add input.
  2. Fill in:
FieldValue
SymbolproductPhoto
DescriptionPicture of the product you want to describe
TypeFile (b64)
Required✓

That single input is all the UI the end-user will see.

3.3. Build the sequence (one simple step)

  1. In Sequence click Add step.
  2. Choose Agent → Photo descriptor.
  3. Paste the prompt below (notice the @productPhoto variable reference – type @ to autocomplete):
Get the photo @productPhoto and:


1. Write a detailed yet concise **Product Description** in plain English.
2. Directly underneath, produce a **GraphQL API call to Shopify** that creates this product, including two option sets:
   • Color (infer values from the image)
   • Size  (Small, Medium, Large)


Use the exact curl format below as a template, replacing the title and option values.


curl -X POST \
https://your-development-store.myshopify.com/admin/api/2025-04/graphql.json \
-H 'Content-Type: application/json' \
-H 'X-Shopify-Access-Token: {access_token}' \
-d '{
  "query": "mutation { productCreate(product: {title: \"Cool socks\", productOptions: [{name: \"Color\", values: [{name: \"Red\"}, {name: \"Blue\"}]}, {name: \"Size\", values: [{name: \"Small\"}, {name: \"Medium\"}, {name: \"Large\"}]}]) { product { id title } userErrors { field message } } }"
}'
  1. Save the flow and then the agent.

4. Test in the built-in debugger

  1. Open the Debugger tab.
  2. Upload any product image (PNG/JPG/WebP).
  3. Click Execute.

The Trace panel shows each step:

  • sequenceAgent → Photo descriptor (generation)
    The final message contains your product description and a ready-made curl command with the correct title and option values.

5. Share or integrate

5.1. One-click share link

  • Go to Run & Share → Share the flow link.
  • Copy the unique URL and hand it to anyone.
  • They’ll see a minimal form (one file-picker) and the same result screen – no login needed.

5.2. Call from your backend

At the bottom of Run & Share you’ll find Run via API.

  • Endpoint – already includes your database-id-hash and flow=describe-product.
  • Modes
    • Execution mode:
      • Synchronous – waits for completion (small jobs).
      • Asynchronous – returns a result_id; fetch the result later (long jobs or batching).
    • Output mode:
      • Stream (default) – chunks partial output for progress bars.
      • Buffer – delivers the whole response in one piece.

Here’s an example synchronous streaming request (replace placeholders):

Terminal window
curl -X POST \
  -H "database-id-hash: 35f5c5b139a6b569d4649b788c1851831eb44d8e32b716b8411ec6431af8121d" \
  -H "Content-Type: application/json" \
  -F flow=describe-product \
  -F execution=synchronous \
  -F output=stream \
  -F productPhoto=@/path/to/photo.jpg \
  https://app.openagentsbuilder.com/exec

The JSON response contains text (the product description) and attachments (if any).

6. Paste into Shopify (or automate it)

The curl snippet returned by the agent still contains:

  • {access_token} – your private-app or custom-app Admin API token.
  • https://your-development-store.myshopify.com – swap for your store domain.

Run the command and the product appears in Shopify Admin → Products with the color- and size-option scaffolding already in place.

Tip: Once you’re comfortable you can remove the curl wrapper from the prompt and instead give the sub-agent the HTTP tool to hit Shopify directly – but keeping it as text first is a safe way to validate output.

7. Next ideas

  • Bulk importer – switch to asynchronous execution and let users drop a ZIP of images.
  • Attribute enrichment – add parallel steps: one agent extracts materials, another suggests price based on similar listings.
  • Language variants – duplicate the description step for DE, FR, ES and merge the results.

That’s it!

You now have a no-code micro-service that turns a photo into a Shopify-ready API call. Duplicate the pattern for invoices, blog covers, or anything else your store needs 🚀