How does BRIX work?

BRIX is a desktop app where each project is a Workspace and each agent is a real terminal running Claude Code, Codex, Antigravity or OpenCode. You lay the terminals out in a Grid or on a Canvas, link them with edges so they can talk, name a lead, launch ready-made squads, watch them in a pixel-art Scene and, on PRO, take the same product into marketing studios.

Updated · by the BRIX team

What are the building blocks of BRIX?

Eight pieces, and each one has a single job. Once you know them, the rest of the app is just combinations.

PieceWhat it isPlan
WorkspaceA project folder plus its terminals. Switching Workspaces never kills a terminal.STARTER
Agent terminalA real terminal running Claude Code, Codex, Antigravity, OpenCode or your shell.STARTER
Grid / CanvasTwo layouts for the same terminals: auto-arranged panes, or a free table with edges.STARTER
EdgeA line between two terminals. It is the permission for them to talk.STARTER
Lead agentThe orchestrator of a linked group. It splits the work and follows up.STARTER
SquadA ready-made team with roles, briefings and edges already drawn.STARTER
SceneA pixel-art office where each agent is a character driven by its real activity.STARTER
StudiosCOPY, Static Ads, Carousel and Video, all reading the same product base.PRO

BRIX does not ship its own model. Every agent runs on the CLI and subscription you already have, so what you pay BRIX for is the environment. See plans.

How does BRIX work, step by step?

  1. 1

    Open a Workspace on your project folder

    Each Workspace points at one folder. That is where every agent you open in it will work. Create more with +, rename with a double click, drag to reorder.

  2. 2

    Launch agent terminals

    One click opens Claude Code, Codex or a plain terminal already inside the folder. Each terminal is a real PTY with your shell, your PATH and your login — not a chat imitation.

  3. 3

    Pick a layout: Grid or Canvas

    In the Grid, panes rearrange themselves every time a terminal opens or closes. On the Canvas, terminals float on a table next to notes, task lists, images and previews.

  4. 4

    Link agents with edges

    Draw an edge between two terminals on the Canvas. From then on they can ask each other things, read each other’s output and split a task — no copy and paste between windows.

  5. 5

    Let a lead orchestrate

    In each linked group, the agent with the most links becomes the orchestrator; you can also set it by hand from the pane menu (Set as orchestrator). In the Grid it takes the left column, full height.

  6. 6

    Or launch a whole squad

    Open BRIX Squads and pick a team. Every agent is born with its role, its briefing and its edges, already linked to whoever it needs to talk to.

  7. 7

    Watch the team in the Scene

    Switch to the Scene and each agent becomes a character at a desk: reading, typing, using a tool or raising a hand when it is waiting for you.

  8. 8

    Take the product into the studios (PRO)

    Switch from CODING to MARKETING, register the product on the dashboard and open COPY, Static Ads, Carousel or Video. Every studio reads the same product base.

What is the difference between the Grid and the Canvas?

They are two layouts of the same terminals, per Workspace. Switching does not restart anything.

GridCanvas
LayoutAutomatic: the grid re-tiles on every new terminalFree: you position, pan and zoom
Manual controlDrag a divider and the app stops re-tiling; Arrange in grid undoes itEverything is manual
Team hierarchyOrchestrator on the left, the others marked “↳ name”Shown by the edges themselves
Other objectsTerminals onlyNotes, task lists, images, previews, reports
Best forTyping in several agents at onceDesigning who talks to whom

There are two more views inside CODING: Scene (the office) and Code (file editor, a form-style .env editor and a diff review against HEAD with commit and push).

How do agents in BRIX talk to each other?

Through a local broker inside the app. Every terminal gets a small CLI called medmaestri, and the edge decides who can reach whom: the edge is the permission. An agent linked to another can run commands like these:

medmaestri list                     # terminals linked to me
medmaestri ask "Backend" "is the /orders contract final?"
medmaestri check "QA" --clean       # read QA's current output
medmaestri send "Frontend" "contract changed, see the note"
medmaestri spawn --name "Tests" --cli codex   # recruit a new linked agent

You do not have to type these yourself. BRIX installs skills that teach Claude Code and Codex to use them, so you ask in plain language (“ask the backend whether the contract is final”) and the agent runs the command.

Agents can also publish to the Canvas: notes, a task list they read and update, images and reports. Each object is born linked to the agent that made it. For the full picture of linking agents, read run multiple Claude Code agents in parallel.

What do the lead agent, squads and Maestro each do?

Three ways to get a team, from most manual to most automatic.

  • Lead agent (orchestrator): in any group of linked terminals, one leads. It breaks the work down, sends each part to the right agent and checks the delivery. It can also recruit more workers on its own with medmaestri spawn — that is Maestro mode on the plans page.
  • Team: a supervisor plus N workers, created in one go from the Scene sidebar or with medmaestri team --count 3, each already briefed and linked.
  • Squads: templates with distinct roles and arbitrary edges. You can duplicate a ready-made one, change roles, or start from scratch and save your own.
  • Maestro, the voice agent (PRO): say “Hey, BRIX” or hold the shortcut and ask out loud — “set up a team to review the checkout”, “what did the agents finish?”. It operates the app and answers by voice.

Which squads come built in?

SquadAgentsWhat it is for
featureArchitect, Backend, Frontend, QAA whole feature, contract first
cacabugRepro, Investigator, FixerBug hunt: reproduce, find the cause, fix
blindagemBRIX Guard, Reviewer, CleanupSecurity audit and diff review
ofertaResearch, Copy, Ads, AuditorOffer launch for the active product
conteudoBrief, Carousel, ReelsOne brief into a carousel and a reels script

More on lead/worker setups in orchestrate a team of AI agents.

What is the Scene in BRIX?

The Scene (labelled Scenario inside the app) is a pixel-art office. Each agent terminal becomes a character that walks to a desk and acts out what the terminal is actually doing. For Claude Code, the activity is read from its local session transcripts; otherwise it follows the terminal’s output.

  • Sitting and reading, typing, using a tool, or idle — taken from the real terminal, not a timer.
  • A raised hand means the agent is waiting for you (a permission prompt, a question).
  • Open a floating window of any agent’s terminal and answer it without leaving the scene.

The point is practical: with five agents running, you see at a glance which one stopped twenty minutes ago asking for permission.

Where do the marketing studios fit in?

This is where BRIX keeps going after the code. On PRO, the top switch goes from CODING to MARKETING, which opens the product dashboard. You register the product once, and its base lives in your folder as Markdown: product.md, audience, offer, brand, FAQ.

StudioWhat it produces
COPYMarket diagnosis, sales angles, copy by objective, a living memory of what worked
Static AdsAd creatives in batches, from 43 templates, 45 archetypes and 15 variation recipes
CarouselTemplate carousels with HTML preview and PNG export
VideoNatural-language editing in HyperFrames: silence cuts, captions, motion

Every studio reads the same base before producing, so the ad repeats what the sales page promised. Each studio’s chat runs on the engine you pick — Claude Code, Codex, Antigravity or OpenCode — or on one of your SuperAgents. More in from code to sales.

What stays on my machine?

  • Everything you produce: Markdown, PNG, MP4 and JSON files in your folders. You can open them without BRIX and version them in git.
  • The agents themselves: they are your CLIs, signed in with your accounts, running locally.
  • The broker: agents talk through a local socket, not a server.
  • Login only validates your plan.

BRIX can also be operated from outside — Claude Desktop, the Codex app or a plain Claude Code terminal — through the BRIX MCP. Ready to try it? Follow getting started.

Frequently asked questions

Does BRIX replace Claude Code or Codex?

No. BRIX runs them: each agent is your own Claude Code, Codex, Antigravity or OpenCode in a real terminal, on the subscription you already pay for. BRIX adds the layout, the links between agents, squads and the studios.

What is an edge in BRIX?

An edge is a line drawn between two terminals on the Canvas. It is the permission for those agents to send prompts to each other and read each other’s output through the local medmaestri CLI.

How is the lead agent chosen?

By default, the terminal with the most links in a connected group is the orchestrator. You can override it from the pane menu with “Set as orchestrator”.

Is Maestro the same as Maestro mode?

No. Maestro mode (STARTER) is a lead agent recruiting and coordinating workers. Maestro (PRO) is the voice agent that operates the whole app when you say “Hey, BRIX”.

Do I need PRO to use squads?

No. The agent environment — Workspaces, Grid, Canvas, edges, squads, the Scene — is in STARTER. PRO adds the marketing studios and the Maestro voice agent.

Which systems does BRIX run on?

macOS on Apple Silicon (M1 or newer, macOS 12+) and Windows 10 (1809+) or 11, on x64 and ARM.

Keep reading