Maestro, the console assistant

Maestro is the AI assistant of the console. It looks up and changes the microfrontends, services, builds, deployments, environments and variables of the project you have open, and searches the Flightroom documentation. It acts as you: with your permissions, on the current project only.

Maestro is available when the project can use at least one AI. Otherwise it is hidden.

Opening it

• As a side panel, from anywhere in the project: ⌘J on macOS, Ctrl+J elsewhere, or the Maestro button in the header.
• As a full page: AI agent, the first entry of the sidebar.

Type in Ask something about the project… and press Send. Suggestions such as Give me an overview of the project or Are there failed builds? Why? get you started. Some pages also offer a button that asks Maestro for you: Explain the failure on a failed build, Analyze the latest deployment, Analyze the conflicts on shared dependencies, Check the variables across environments.

Next to the message box you can pick the AI and the Reasoning effort. Every answer shows the tokens it used. Conversations are saved per project: Recent conversations reopens them, New conversation starts a fresh one.

Confirming actions

Reads and ordinary writes run directly. Actions that are destructive or that hand out access stop and wait for you. The panel shows a Confirm this action card, and nothing happens until you press Confirm; Cancel drops it.

Actions that need confirmation include:

• deleting anything, deploying, rolling back (a redeploy of a previous snapshot, which does not touch Kubernetes: see Rolling back a Kubernetes deploy), committing to a repository;
• changing a member's role, removing a member, revoking an API key or an AI client connection;
• inviting to a project or an organization, resending an invitation, creating an API key;
• discarding an AI change, setting canary users.

The card names what the action acts on (users, environments, microfrontends with version and host, services, domains, the AI change with its request and files) instead of raw ids. A deploy card shows, for each environment, which versions, hosts and variables change compared with the active deployment. Input sent by Maestro shows exactly what the model asked for.

A card can be answered once and expires after 15 minutes. After that, ask again.

Starting AI code changes

The New AI job button in the panel starts an agent on microfrontends and services you pick, across domains. Jobs open as tabs in Maestro, where you follow them, chat with the agent and apply the proposals. See Change code with AI.

What Maestro does not do

Some things stay in the console pages on purpose, because their input or output is a secret that would travel to the AI provider:

• storages;
• code repository connections;
• your user profile;
• bundle upload.

When Maestro creates an API key, the new key is shown in the panel only and never sent to the model. Build logs it reads are redacted first: known token formats, credentials in URLs, variables named like secrets, private keys and similar values are blanked.

Permissions

Maestro runs every tool through the same checks as the console. On a project where you are a Viewer it cannot change anything. Member management (inviting, changing roles, removing) takes a project owner or an administrator of the organization, as in the console pages.

Audit

Every answer and every write is logged by the server:

Audit tag	Logged for	Contains
assistant_usage	Every answer	The user, the project, the AI, the effort and the token counts
assistant_tool_call	Every write, confirmed or refused	The tool, the project or organization, the user, the outcome, and the arguments reduced to ids and short identifying values. Never free text or secrets

Questions count against ASSISTANT_RATE_LIMIT_MAX per user per minute (default 20). Past it, Maestro answers Too many questions in a short time. Wait a minute and try again.