Flow Generation (Beta)
Learn how to use AI to help you create pipelines on the Digibee Integration Platform.
The Flow Generation is currently in beta phase. Learn more about the Beta Program.
Flow Generation is a key feature of our AI Pair Programmer approach on the Digibee Integration Platform. Simply provide a prompt, and the AI will generate a draft pipeline for you, helping you create parts of integration flows or entire pipelines more quickly, tailored to your specific needs.
To access the Flow Generation and use the features presented in this article, you need the ai-assistant-viewer system role, which is available by default in the developers group. Learn more in the documentation about Roles.
How to generate a pipeline with Flow Generation
Generating the pipeline
On the Build page, click Create in the upper right corner and select Pipeline.
In the pipeline, click the Sparkle icon (next to the Settings button) to start using the AI Assistant.
Select the Flow Generation from the options in the chat.
Learn more about the JOLT Generation and Docs Q&A options.
Describe in the chat what you need for your integration. For more information on how to write effective prompts, read the How to write effective prompts section.
Reviewing the pipeline
The Flow Generation will create part of a pipeline or an entire pipeline based on your request. It will show you the flow tree of the generated pipeline and provide the documentation on how it works.
The Flow Generation only creates the pipeline structure, meaning that connector parameters are not generated by the AI yet — only the Step Name is provided.
After reviewing the generated flow, you have the following options:
Request changes: analyze the pipeline and request changes, such as adjusting the flow, removing connectors, or adding any missing connectors. When requesting changes, be sure to follow the best practices outlined in How to write effective prompts.
Accept the suggested flow: if you’re satisfied with the flow, click the check icon to accept it. The information will be saved in the chat while the chat history remains available.
Discard the suggested flow: to reject the AI’s suggestions, simply click the trash icon.
Once you are satisfied with the result, you can:
Copy the flow and paste it in the Canvas (or in other pipelines/capsules).
Insert on Canvas to place the flow directly into the Canvas.
If you close the AI Assistant, you can still access the chat history while building the pipeline, as long as you don’t refresh the page. To clear all the chat history, click the trash icon in the upper right corner.
The chat history is reset every time you leave the pipeline or refresh the page.
How to write effective prompts
OpenAI’s documentation provides a comprehensive guide on creating more precise prompts through prompt engineering. Access the documentation for more details.
For Flow Generation, here are some additional tips to help you get the most out of our AI Assistant. Keep these in mind when formulating your prompt:
Be specific about the structure of the flow you want to generate.
Write clear and direct instructions that describe exactly the part of the pipeline or the entire pipeline that you want the AI to generate.
Use an organized structure and separate ideas with commas or paragraphs to ensure clarity.
Mention specific connectors if you already know which ones to use.
Break the prompt into parts if it’s complex and ask the AI to make adjustments if necessary.
Below are some dos and don’ts to consider when creating your prompt, with examples provided for better understanding.
Prompts with specific connectors
Do:
Provide clear, objective examples. If you already know the connectors that should be included in your integration, mention them.
✅ I want to create a pipeline that starts by using the REST V2 connector to make a request. After that, it will use a Choice connector to evaluate the API response, followed by a Log to check the result. If the request is successful, the pipeline should use the JSON Generator to create the output message. If there’s an error, the Email Connector V2 will be used to send a notification, ending the flow.
Don’t:
Be vague or leave out important details. Avoid describing only general steps without mentioning specific connectors or outcomes.
❌ I want to create a pipeline that makes an API request, checks the response, and then sends an email if something goes wrong.
Prompts with endpoint types
Do:
Instead of listing all the connectors, you can mention the types of endpoints (for example, DB, REST API) for each side of the integration and let the generator choose the most suitable flow.
✅ Connect my SAP ERP system to a database and a REST API.
Don’t:
Be too generic or omit essential details that could make it difficult for the generator to create the appropriate flow.
❌ Connect my ERP system to a database and a REST API.
Prompts without mentioning connectors
Do:
Even if you are unsure about the specific connectors to use, you can still create a prompt that effectively communicates your goals. Ensure your prompt describes the overall objective and provides sufficient details for the AI to understand your requirements.
✅ I need to create a pipeline integration that reads data from a stream database, publishes an event, and writes the data to another database. It starts by connecting to the stream database to read the necessary data. The data is then published as an event to notify other systems or services. Finally, the data is written to another database for storage or further processing.
Don’t:
Be overly vague when describing your integration’s goal. Avoid leaving out important information that helps the AI understand the complete process.
❌ I need to create a pipeline that reads data, publishes it, and writes it somewhere else.
Limitations
The Flow Generation has some known limitations:
Large pipelines: the AI can generate hallucinations when handling excessively large pipelines.
Ignorance of context: the AI doesn’t recognize the context of an existing pipeline already on the Canvas. It generates new content without associating it with the current pipeline.
Parameters handling: the AI doesn’t capture parameters. Its main focus is on the logical structure and the connections between the connectors.
Last updated
