Cross-Agent Mentions

Cross-Agent Mentions let you pull another agent into your current conversation without leaving it. Type @ in any AI input, pick an agent, and Maestro forwards the relevant slice of your chat to that agent, runs it in the background, and streams its answer back inline - stamped with who replied. It is the lightweight cousin of Group Chat: no moderator, no shared room, no ceremony. Just a quick “what does the backend agent think about this?” from wherever you already are.

​

When to Use It

• Second opinion - “@Reviewer does this migration look safe?” without copy-pasting your thread into another agent.
• Cross-project context - Ask the agent that owns the backend repo a question while you work in the frontend one.
• Specialist consult - Route a security or performance question to the agent that has that codebase loaded.
• Quick fan-out - Mention several agents (or a whole group) in one message and let them answer in parallel.

​

Mentioning an Agent

1. In the AI input, type @. The mention picker opens with two categories: Files and Agents (agents and groups you can reach from this agent).
2. Keep typing to filter, or use the arrow keys. Pick an agent to insert an @name token, which renders as a chip.
3. Write your question as usual and send. The @name stays in your message so the consulted agent knows it was addressed.

The picker uses a single @ for everything. Maestro tells files and agents apart by shape: a path-like body such as @src/app.ts or @notes.md is a file reference, while a bare word like @codex is matched against your live roster of agents and groups. A @word that names nothing stays plain text.

​

What Happens When You Send

The consultation is non-blocking and isolated:

• Maestro forwards a window of your current tab’s transcript plus your prompt to the target agent.
• The target runs as a fresh, ephemeral process - it is not injected into that agent’s own live chat, so the consultation never pollutes the other agent’s main conversation.
• The answer is still kept. Maestro writes it to a dedicated consult tab on the target agent, labeled with who asked (↩ YourAgent), so the target has a durable record of what it was consulted about. That agent’s History also logs a “Consulted by YourAgent” entry, so it remembers who reached out.
• Continuity per thread. Ask the same agent again from the same tab and Maestro resumes its consult session, so it carries forward your earlier consults from that thread. A mention from a different tab starts a fresh consult in its own tab.
• Your chat is never blocked. A small pill at the top of the input shows in-flight consultations (each agent’s name and elapsed seconds); click it to expand the list. Keep typing while you wait.
• When the target finishes, its reply streams back inline into the chat you are already in, attributed to the agent that answered.

Every consulted reply lands in a tinted bubble topped by an attribution header: the answering agent’s name, its provider, and its session id. That header is what tells replies apart when several agents answer at once, and it does double duty as a jump control. Click the agent name (or the jump button on the right) to open that agent’s consult tab in the Left Bar, where the full exchange is kept, so you can continue the thread in its own context; click the session id to copy it. While a reply is still streaming the header shows a spinner, and a consult that failed tints the header red. Mention several agents in one message and each runs independently and concurrently, so a fan-out returns as fast as the slowest agent, not the sum of them.

​

Who answers: your agent, the mentioned agents, or both

Whether your current agent also answers depends on where the mention sits:

• Start the message with an @agent mention (@Backend does this look right?) and the message is treated as addressed to the mentioned agent(s) only. Your current agent stays quiet; you still see your message in the chat as the anchor for the replies that stream back.
• Put the mention later in the sentence (does this look right to @Backend?) and your current agent answers too, with the consulted agent’s reply arriving alongside it. Use this when you want both perspectives.

A leading @file reference (@src/app.ts what does this do?) is a question for your current agent about that file, so it does not count as addressing another agent.

​

Controlling How Much Context You Share

By default Maestro forwards your entire current transcript so the other agent has the full picture. When that is more than you want to share, add a natural-language hint to your message and Maestro narrows the slice automatically: An explicit count always wins over a softer hint, and the hint is read from your prose only - the @name token itself is ignored when Maestro decides the window.

​

Mentioning a Group

Mention a group by name (@Backend-Team) to consult every agent in it at once. Group mentions expand to each non-terminal member and run as independent consultations. If you mention both a group and an agent that belongs to it, that agent is still consulted only once. Groups sort above individual agents in the picker, so a name that matches both surfaces the group first.

​

Cross-Agent Mentions vs Group Chat

Both let agents talk to each other, but they solve different problems: Reach for a mention when you just need an answer; open Group Chat when you need agents to deliberate together over several rounds.