AI Copilot
Using the AI Copilot
The AI Copilot is a chat-based interface that is located in the right sidebar of your Flow Editor. It helps you build and maintain flows, and once you enter a prompt, it provides rich answers in formats such as text, flow snippets, code snippets, tables, images, and diagrams.
Access the AI Copilot in the right sidebar of the Flow Editor.
The AI Copilot provides conditional actions through buttons, such as to apply changes immediately or import a duplicate of the generated flow.
The AI Copilot is designed to assist you with the following actions:
Creating Flows:
Describe what you need, and the Copilot will generate a new flow by selecting the best nodes, writing code for function nodes, and integrating community and qibb nodes to jumpstart your project.Editing Flows:
Easily update your existing flows using conversational commands. The Copilot helps you refine your flows, making them more organized, resilient, and secure.Troubleshooting & Debugging:
Get assistance in diagnosing issues by reviewing the log and run history, with suggestions for quick fixes.Explaining Flows in Text and Diagrams:
Receive clear, detailed descriptions and step-by-step breakdowns of your flows to boost understanding and speed up onboarding.Answering Questions & Helping with Best Practices:
Answers your questions by browsing qibb’s product documentation. The Copilot provides step-by-step instructions based on our public documentation, helping users learn, optimize flows, and better understand the product.
FAQ
When is the AI Copilot available?
qibb’s AI Copilot is now live in the demo environment and will be gradually rolled out to all existing customers. The rollout will commence in our SaaS environments and subsequently extend to our dedicated PaaS environments in March and April. Please note that the availability of this feature may vary based on individual account configurations and technical considerations.
Which Subscription Tier includes this feature?
This feature is available for all subscription tiers at no extra cost, although usage limits will vary depending on your subscription level.
How reliable are the suggestions provided by the AI Copilot?
Our AI Copilot leverages advanced large language models to generate suggestions designed to help you build and optimize your workflows.
However, there are limitations and it's important to review each response carefully before implementation. Some key points to consider:
Context Matters: The quality of suggestions depends heavily on the clarity and context of your input.
Manual Adjustments: While the AI provides a strong starting point, additional manual configuration of nodes may be required.
Occasional Inaccuracies: Like any AI system, the Copilot may sometimes produce outputs that do not fully match your intent—these are often referred to as "hallucinations."
Verification Recommended: We advise verifying critical details and treating the outputs as a helpful guide rather than definitive answers.
Feature Limitations: Note that certain features and actions may have limitations. Please refer to the product documentation for a comprehensive list of current feature limitations.
We are always working to make the AI Copilot better and more accurate, and we'll keep updating it regularly.
How can I improve the accuracy of the AI Copilot's responses?
To enhance the precision and relevance of the AI Copilot’s output, consider the following tips:
Be Specific: Provide clear, detailed, and specific instructions in your prompt. Incorporate relevant context or examples to help the AI understand your requirements better.
Iterative Refinement: Request minor changes or flow segments step by step, following an iterative approach.
Focus the Scope: Use the option to toggle different data sources (for example, limit the context to the flow selection) to help the Copilot focus on the most relevant information.
Tip: Use Comment Nodes for Copilot Guidance
Insert comment nodes in your flow to add extra documentation for the Copilot. For instance, you can include extra system instructions or set a style guide for the Copilot on each flow tab. This information will be utilized as long as you share the deployed flow for context.
Can I bring my own AI Provider Account?
Currently, this feature does not support the use of a custom or third-party AI provider account.
What AI model is used?
We currently utilize multiple large language models (LLMs), including those provided by OpenAI, to power the AI Copilot.
Is my data safe when using the AI Copilot?
The Copilot only accesses data that you have explicitly allowed, and sensitive information is never stored or used for training. You remain in full control of your data at all times.
What Data Sources is the AI Copilot accessing?
The AI Copilot can access a variety of data sources on-demand. Users have full control over which data points are available during their session.
Options menu with data controls
The data sources include:
Flow Resources:
Deployed Flow: The active flow currently running. You can choose to limit sharing only the currently selected nodes of the flow.
Context Variables: Only the variable names are accessed (values are not read).
Secret Variables: Only the variable names are accessed (values remain confidential).
On-demand Knowledge Resources:
Nodes from qibb’s Catalog
Flow Templates from qibb’s Catalog
Public Documentation of qibb
Platform Monitoring Data:
Log History: Logs of the last 30 minutes related to the flow.
Run History: Runs of the last 30 minutes related of the flow.
What diagram formats can be generated?
The AI Copilot can generate visualizations in several diagram formats based on your prompts. This includes visualizations for flow logic and other specific data needs.
Supported diagram formats include:
Sequence Diagram
Flowchart Diagram (experimental)
Pie Chart Diagram (experimental)
Note: The experimental formats may not render consistently in every scenario.
Can I restrict access for my users to the AI Copilot?
Yes, you can control access using a feature toggle available at the app level during deployment or upgrade procedures initiated from the Portal. If the feature is enabled, users with the Edit or View app role will have access to the AI Copilot in the Flow Editor.
Current Limitations
Some features and actions in the AI Copilot have known limitations, but we're working to improve the product and value your feedback. Please review the current limitations below:
General AI Limitations
The accuracy and quality of outputs depend on various factors, and results may vary.
Always review generated flows and code before implementing them. For additional tips on improving outcomes, please refer to the AI Copilot FAQ.
No Web Search
The AI Copilot currently does not access any external web resources or offer web search functionality, so please include all necessary information directly in the chat.
Chat History
Chat history is not saved and will be cleared if you refresh your browser.
The Copilot only considers the latest 10 user messages and responses when processing new prompts.
Diagram Rendering
Diagram formats other than Sequence Diagrams are experimental and may occasionally not render correctly.
Manual Node Installation
The AI Copilot does not automatically install missing nodes. You will need to install them manually. Please refer to our documentation for guidance:
Node Properties Configuration and Data Mapping
Some node configurations may not always be considered or applied by the AI Copilot. This can be applicable for error handling, input/output mapping, security fields for API Nodes and Authentication Nodes.
Occasionally, nodes generated from qibb’s catalog may have incorrect operation selections or parameter configurations that do not match the node schema; in such cases, invalid settings are either ignored or appear empty when you open the node properties.
Occasionally, data mapping between nodes is not implemented correctly, causing generated nodes to reference default variable paths (such as msg.payload) even when an alternative variable path is desired. Review according variables to match your desired mapping.
Working with Flows
The Copilot only considers one single flow tab at a time—specifically, the currently active tab when a prompt is sent. Utilizing multiple flow tabs for context during a single prompt request is not yet supported.
The Copilot may not generate or interpret subflows correctly. We recommend creating flows at a flat level with the Copilot and then manually wrapping the logic into subflows afterward.
Working with JSONata Expressions
JSONata expressions generated by the Copilot may occasionally contain syntax errors, especially within the Body Payloads or Parameters of API nodes from qibb’s Catalog.
Working with Function Nodes
The “Import” syntax for external modules may not correctly be generated within a function node. Please refer to How to use external modules in a function node for guidance.
These limitations are part of our ongoing efforts to improve the product. For future updates, check the release notes of Flow Releases and Platform Releases.