Build with AI overview
Anyone in a project can describe the app or the AI agent they need in their own words. Flightroom's AI writes the code in your own Git repositories (GitHub, GitLab or Azure DevOps), starting from Flightroom templates, and pushes it to a branch of its own. Engineers review the change where they already review code, and from there it ships like anything else in the project.
The AI never writes to your base branch by itself. It proposes, and a person applies, opens a pull request or discards.
Apps and agents in console terms
The console has no entity called "app" or "agent". What you describe maps to the building blocks Flightroom already manages:
| What you ask for | What it is in the console |
|---|---|
| The web interface of an app | A microfrontend. Its Kind is Microfrontend (served by Flightroom) or Frontend (served by its Docker container) |
| The backend or API of an app | A service, listed under Backend services |
| An app as a whole | A functional domain (Domains in the sidebar), which groups the microfrontends and services of one business capability |
| An AI agent | A service whose code the AI writes to call an AI model and your systems |
An agent has no separate runtime. It is a service like any other: it lives in its repository, goes through your CI, gets versioned and is deployed the same way.
The rest of this section uses the console terms when it describes what to click.
What the AI does, and what it leaves to you
- It works on a clone of the repository, never on the repository itself.
- It commits its proposal on a new branch,
ai/<slug>-<timestamp>, and pushes it. The branch it started from is not touched. - It does not install, build or test anything. Your CI checks the code, as it does for any other branch.
- It cannot read files that usually hold credentials, and cannot change CI configuration or lockfiles. See Guardrails.
Once the proposal is ready you choose: Apply it to the base branch, Open a pull request, or Discard it. See Change code with AI.
Ways to build
| Starting point | Where | Page |
|---|---|---|
| A new app from nothing | An empty functional domain, Edit with AI tab | Generate a domain |
| A change to a microfrontend, a service or a whole domain | Its Edit with AI tab | Change code with AI |
| Questions and actions on the project, in a chat | Maestro, ⌘J / Ctrl+J or AI agent in the sidebar | Console assistant |
| Your own AI client (Claude, ChatGPT, Gemini, Cursor…) | The Flightroom MCP server, on installations that turn it on | MCP server |
Set it up once
An administrator prepares the ground; after that, project members build.
-
Add an AI to the organization. An owner or an admin of the organization opens Organization settings, tab AI, and adds one with Add AI: Claude (Anthropic, Azure AI Foundry or AWS Bedrock), Gemini or GPT on Azure OpenAI. Every project of the organization can use it unless restricted. See AI models.
-
Connect a code repository provider to the project. The AI works in repositories reached through a code repository connection: GitHub, GitLab or Azure DevOps.
-
Optionally, write AI context. Free markdown notes for AI agents (business domain, conventions, rules), up to 20,000 characters each:
- for the whole project, in Settings → General → Project AI context;
- for a functional domain, in the AI context field of the domain form;
- for a microfrontend or a service, in its AI context tab.
The agent always receives the context of what it works on, of its domain and of the project.
When a project has no AI available, the AI features are hidden or report that they are not enabled, and nothing is sent to any provider.
Who can do what
| Action | Who |
|---|---|
| Add, edit or delete the AIs of the organization, and choose which AIs each project can use | Organization Owner or Admin |
| See the AI change requests of a project | Every member of the project |
| Ask for a change, apply it, open its pull request, discard it | Project Admin and Editor, and organization Owners and Admins. Not Viewer |
| Use Maestro | Every member; it acts with the member's own permissions, so a Viewer cannot change anything through it |
| Connect an AI client over MCP with changes allowed | Members who can write on the projects they share; on a project where they are Viewer the client stays read-only |
The AI features check the project role: a Viewer can read AI change requests but not start, apply or discard one. Elsewhere in the console the project role is still a label; see Project members and roles.
From the branch to production
The AI's work ends in your repository. From there the usual path applies: CI builds the branch you merged or applied to, and you deploy the result to an environment. To run your microfrontends and services on your own cluster, see Deploy to Kubernetes.
Where to go next
- Generate a domain: from an empty domain to a first version
- Change code with AI: proposals, chat, apply, pull request, discard
- Guardrails: limits and protections, for IT and security reviewers
- AI models: connecting your AI providers
- Console assistant: Maestro
- MCP server: using Flightroom from your own AI client