Run several AI coding agents like Claude Code or Codex at once and it quickly becomes hard to tell what is happening in which terminal. Maestri is a desktop app that tries to solve that confusion with a single infinite canvas. Terminals are treated as freely positioned nodes, and dragging a connection between two of them lets the agents exchange prompts directly over PTY and work together. It is built by Evert Junior, an individual developer based in Brazil. No account is required, there is zero telemetry, and all data stays on your own machine. macOS and Windows are supported.
Key Features
- Connect agents with a line and let them collaborate: Draw a connection between terminal nodes on the canvas and the agents exchange prompts directly through a PTY (pseudo-terminal). There is no need to copy one agent’s output into another by hand
- See everything at once on an infinite canvas: Place terminals, sticky notes, a file tree, diff views and device portals wherever you like. Pan and zoom are unlimited, so you can build a layout per project
- Ombro (on-device AI assistant): Built on Apple Foundation Models, it summarizes the work agents did while you were away. Processing is 100% on-device
- Roles (reusable custom instructions): Save agent configurations with custom instructions ─ a “reviewer” or an “implementer”, for example ─ and call them up when needed, turning per-project setups into templates
- Remote connections and session persistence: Connect over SSH, Docker or a sandbox. tmux keeps agent sessions alive after the app is closed, so long-running remote work is not interrupted
- Device portals: Embed an iPhone simulator, an Android emulator or a browser directly on the canvas, so you can check what an agent produced while giving it the next instruction
Pricing
| Plan | Price | What you get |
|---|---|---|
| Free | $0 | 1 workspace. All core features including the infinite canvas, agent connections, roles, notes and Ombro |
| Pro | $18 (one-time) | Unlimited workspaces, use on up to 2 Macs, workspace navigation shortcuts. 7-day free trial |
Pricing is as of August 2026. Check the official site for the latest information.
A one-time perpetual license rather than a subscription is unusual for a tool in this space. The free plan has almost no feature restrictions ─ the practical difference is the number of workspaces. A natural path is to start free and consider Pro once you are juggling several projects in parallel.
Pros & Cons
✅ Pros
- A UX with few parallels: connect agents with a line and they collaborate
- No account, zero telemetry, all data stored locally ─ a low psychological barrier when working with production code
- Native implementation in Swift/Metal keeps it lightweight
- SSH, Docker and sandbox connections plus session persistence make it usable for remote development
- Pro is a $18 one-time purchase, so no monthly fees accumulate
⚠️ Cons
- The Pro license is limited to 2 devices
- MCP (Model Context Protocol) is not supported
- Only macOS and Windows are supported ─ there is no Linux build
- It is purely an orchestration layer; agent CLIs such as Claude Code must be installed separately
- Because the design is local-only, there is no cloud sync or multi-person collaborative editing
Comparison with Similar Services
| Item | Maestri | Conductor | Warp | Superset |
|---|---|---|---|---|
| Core idea | Spatial layout on an infinite canvas | Agent isolation via Git worktrees | Reinventing the terminal itself | Worktree isolation + broad agent support |
| Direct agent-to-agent link | Yes (connect and converse) | No | Mainly parallel execution | Mainly parallel execution |
| Supported agents | Claude Code / Codex / OpenCode and others | Claude Code / Codex | Broad support | Broad support |
| Supported OS | macOS / Windows | macOS | Cross-platform | Cross-platform |
| Where data lives | Fully local (no telemetry) | Local | Cloud integration available | Mostly local |
Conductor goes in the direction of “isolate each agent in its own Git worktree and channel it into a review-and-merge flow”, while Maestri chooses to “lay several agents out spatially and let them talk to each other when needed”. If you want an orderly review flow, Conductor fits; if you want to run many agents exploratively and watch the whole picture, Maestri fits.
Who Is It For
- Developers running several instances of Claude Code or Codex at once who struggle to keep track of progress
- People working with agents in remote or container environments over SSH or Docker who want them visible on one screen
- Anyone who, because they handle production code, prefers a local-only tool with no telemetry or cloud transmission
- Those who would rather not add another monthly subscription and prefer a one-time license
- Developers who want to check mobile app or web output while directing agents
Summary
Maestri answers the relatively new problem of running AI coding agents in parallel with a spatial UI built around an infinite canvas. The ability to connect agents with a line and have them talk directly has few parallels, and it is worth trying for developers who use multiple agents daily. At the same time there are constraints ─ no MCP support, no Linux build, and the two-device limit on Pro ─ and it assumes you bring your own agent CLIs. Since the free plan already includes nearly all core features, the sensible first step is to try it within a single workspace.