馃摑 AI Catalog: Feedback issue
Welcome to the AI Catalog (Experimental) Feedback Issue
The purpose of this feedback issue is to collect your experiences with the AI Catalog, including the creation and retrieval of agents and flows. Our goal with this issue is to understand your experience with the AI Catalog and related workflows.
What is the AI Catalog?
The AI Catalog is a central registry where you can create, discover, and manage agents and flows within the Duo Agent Platform, located at the instance level via the Explore menu. It provides:
- Create: Build new agents and flows for your projects
- Discover: Browse, search, and understand what AI capabilities exist
- Enable agents and flows in your projects to use them across the GitLab Duo Agent Platform
What's the difference between agents and flows?
Agents are AI-powered assistants that help you accomplish specific tasks and answer complex questions. There are three types in GitLab:
- Foundational agents: Pre-built, production-ready agents created by GitLab for common workflows. These agents come with specialized expertise and tools for specific domains. Foundational agents are turned on by default, so you can start using them with GitLab Duo Chat.
- Custom agents: Agents you create and configure for your team's specific needs. You define their behavior through system prompts and choose what tools they can access. Custom agents are ideal when you need specialized workflows that aren't covered by foundational agents. To interact with a custom agent, enable it in a group or project to use it with Chat.
- External agents: Integrate with AI model providers outside GitLab. Use external agents to allow model providers like Claude to operate in GitLab. You can trigger an external agent directly from a discussion, issue, or merge request.
Flows are a combination of one or more agents working together to solve a complex problem. There are two types:
- Foundational flows: Pre-built, production-ready workflows created by GitLab for common development tasks
- Custom flows: Workflows you create to automate your team's specific processes. You define the workflow steps and agents, and define triggers to control when the flow runs.
Flows are available in IDEs and the GitLab UI. In the UI, they run directly with Runner job execution, helping you automate common development tasks without needing to leave your browser.
What is a trigger?
A trigger determines when a flow runs. You specify the service account that runs the flow, and which conditions make the flow run. For example, you can specify flows to be triggered when you mention a service account in a discussion, or when you assign the service account as a reviewer.
You can configure a flow to run when:
- Mention: The service account user is mentioned in a comment on an issue or merge request
- Assign: The service account is assigned to an issue or merge request
- Assign reviewer: The service account is assigned as a reviewer on a merge request
Examples of agents and flows
Below are examples of team member created agents and flows for inspiration. Feel free to add to the list!
- Flow: VueJS Unit Test Writer
- Flow: Issue triage
New features with 18.7
- Version pinning
- GitLab managed external agents
- Group-level Agent and Flow pages
- Agent Identity (for easier flow trigger setup)
- Reporting (and deleting) of Malicious agents
- Audit Events for Agents and flows
- UX improvements
馃幆 Feedback we're especially interested in
We are actively working to mature the AI catalog, agents and flows features as well as the overall user experience and end to end workflows. We need your help in identifying:
- Abrasion: What made it hard for you to engage with these features?
- Discovery: Was it hard to find something or how to complete something?
- Missing capabilities: What tasks do you want to use DAP for, but can't currently?
馃摑 How to give feedback
-
Check existing feedback: Review threads below to see if your issue is already reported. Add a
馃憤 or comment to show support. - Start a new thread: Use a descriptive title so team members scanning open threads can quickly identify the focus of your thread.
- Include context: Tell us what you were trying to do, why and what happened. Include screenshots where relevant.
馃 What you can expect from us
- We will read all feedback while this issue remains open
- We will prioritize fixes based on feedback patterns
- We will create issues for reproducible problems with severity/priority labels
- We may reach out for clarification on complex issues