Skip to content

AI Composer

The AI Composer lets you describe an orchestration in plain English and have DIBOP generate the steps automatically. Instead of manually configuring each step, you tell the AI what you want to achieve and it creates a working orchestration for you.

How It Works

  1. You describe what you want the orchestration to do in natural language
  2. You select which connected systems should be involved
  3. The AI analyses your description, the selected systems' APIs, and the Canonical Data Model
  4. It generates a complete orchestration with properly configured steps and parameter mappings
  5. You review, edit if needed, and activate

The AI Composer uses advanced language models to understand your intent and translate it into a working workflow.

Using the AI Composer

Step 1: Open the Composer

Navigate to BUILD > Orchestration Composer in the sidebar. You will see the Composer interface with a text input area and system selection panel.

Step 2: Describe Your Orchestration

Write a plain-English description of what you want. Be specific about:

  • What data to fetch and from where
  • What transformations to apply
  • Where to send the results
  • How to handle errors

Good description:

"Fetch all vehicles from Mercedes-Benz OneAPI that were updated in the last 24 hours. For each vehicle, decode the VIN using NHTSA to get the full specifications. Map the combined data to the Vehicle canonical domain and write it to our DMS. If the DMS write fails, log the vehicle to a dead letter queue."

Too vague:

"Sync vehicles between systems"

Step 3: Select Systems

Choose the connected systems that should be involved in the orchestration. The AI uses the selected systems' registered API operations to generate appropriate steps.

Be Specific About Systems

The more specific you are about which systems and operations to use, the better the AI's output will be. Mentioning specific API operations (like "decode VIN" or "search vehicles") helps the AI choose the right endpoints.

Step 4: Generate

Click Generate. The AI will process your description and produce an orchestration with:

  • A name and description
  • Input parameters (if needed)
  • Ordered steps with configured operations and parameter mappings
  • Error handling where appropriate

Generation typically takes 5-15 seconds, depending on the complexity of the orchestration.

Step 5: Review the Output

The generated orchestration is displayed in the editor. Review each step:

  • Step names: Are they descriptive and clear?
  • Operations: Are the correct API operations selected?
  • Parameter mappings: Do the data references look correct?
  • Error handling: Is it appropriate for your use case?

You can edit any aspect of the generated orchestration before activating it.

Tips for Better Results

Be Explicit About Data Flow

Tell the AI exactly what fields you need and where they should go:

"Take the vin, make, model, and year from the OEM response and write them to the DMS as stock_vin, vehicle_make, vehicle_model, and model_year."

Mention Error Handling

If you need specific error handling, include it in your description:

"If the VIN decode fails, skip that vehicle and continue with the next one. After processing all vehicles, send a summary email with the count of successes and failures."

Reference the Canonical Data Model

If you want the orchestration to use CDM fields, mention it:

"Map the response to the Vehicle canonical domain before writing to the target system."

Specify Conditions

If your workflow needs branching, describe the conditions:

"If the vehicle's model year is 2024 or newer, add it to the 'new inventory' category. Otherwise, add it to 'used inventory'."

Editing AI-Generated Orchestrations

The AI Composer generates a starting point -- you can and should refine it. Common edits include:

Edit Why
Rename steps Make step names more descriptive for your team
Adjust parameter mappings The AI may not know your exact field names
Add error handling Add retries or fallbacks the AI did not include
Remove unnecessary steps The AI may add steps you do not need
Add steps Include additional API calls or transforms the AI missed

You Own the Result

The generated orchestration is fully editable. It is the same format as manually-built orchestrations, so you can switch between the Composer, Visual Builder, and manual editing freely.

Standalone Test Fire

The AI Composer includes a Test Fire feature that lets you test an individual step without running the entire orchestration.

How to Test Fire

  1. After generating an orchestration, click on any API Call step
  2. Click Test Fire
  3. Enter sample parameter values
  4. Click Execute
  5. View the raw API response

Test Fire calls the real external API, so you see actual responses. This is useful for:

  • Verifying that the AI selected the correct API operation
  • Checking the response structure to confirm parameter mappings
  • Testing with different input values

Live API Calls

Test Fire makes real API calls to the external system. Be cautious with operations that modify data (POST, PUT, DELETE) -- test with safe or sandbox data.

Limitations

The AI Composer is powerful but has some limitations:

  • Complex business logic: Very intricate conditional logic or multi-level nesting may need manual refinement
  • Custom transforms: Complex data transformations (aggregations, pivots) may need manual adjustment
  • New or unusual APIs: If a connector was recently added and the AI has not seen its response format before, mappings may need correction
  • Enterprise-specific conventions: The AI does not know your internal naming conventions or business rules unless you include them in the description

Regenerating

If the initial output is not what you expected:

  1. Refine your description to be more specific
  2. Click Regenerate
  3. The AI creates a new orchestration from scratch (the previous version is not modified)

You can generate multiple versions and compare them before choosing one to activate.

Next Steps