Armadai vs AgentGrid: two canvases for coding agents

AgentGrid and Armadai both put coding agents on an infinite canvas, and they differ most in where they run, which agents they drive, and what one canvas holds. AgentGrid runs on macOS, Windows and Linux, drives nine agent tools including Claude Code, Codex, Cursor and OpenCode, and has a Free plan for three projects. Armadai is a native Mac app for Claude Code and Codex that puts several projects on one canvas, with task boards and a code editor beside the agents, and it is a subscription with no free plan.

The comparison in one table

A canvas means something different in each app

In AgentGrid, the top-level unit is a space. Each space appears as a tab, owns one canvas, and is bound to one project folder, so new agents and terminals start in that directory. You can point several spaces at the same repository, for example one canvas for building and another for review. Activity dots on the tabs show which spaces have work running while you look at another one, and a workspace menu lists your projects and agents together.

In Armadai, each project folder becomes a workspace, and several workspaces sit on the same canvas. You zoom out to see every project and zoom in to work in one. If you prefer tabs, Tab View gives each workspace a tab with a canvas of its own, and you can keep separate canvases, such as one for your product and one for client work. A sidebar lists agent sessions across all of them as working, idle or waiting for approval, and you can answer a waiting agent from there.

The practical difference is whether different projects share a canvas. In AgentGrid, each project has its own canvas and you move between them by tab. In Armadai, you can lay out several projects on one canvas, or switch to tabs if you would rather see one at a time.

What AgentGrid has that Armadai's homepage does not list

• Windows and Linux builds. AgentGrid's download page offers Windows installers for x64 and Arm, and Linux AppImage and Debian packages for x86_64 and ARM64.
• A Free plan. Free covers three projects with up to 10 parallel agents each, running locally.
• Nine agent tools. As well as Claude Code and Codex, AgentGrid drives Cursor, OpenCode, Antigravity, Kimi, Grok, Devin and Pi.
• A different agent and model for each worker. A master starts workers for focused jobs, and each can run any installed agent with its own model, so one team can mix a Claude builder, a Codex reviewer and an OpenCode validator.
• Usage tracking and model switching. The title bar shows usage for the agents you have installed, with reset times. You can switch models in a conversation and keep its context, and Auto-switch hands a turn to your next starred model when one hits its limit.
• Voice calls with an agent. You can start a call from an agent pane and talk through the work while it runs.
• Cloud sync and cloud agents on Pro. Pro adds unlimited canvases, sync across machines, and cloud as well as local agents.
• A phone app in beta. AgentGrid for iPhone is in TestFlight, with invites sent by hand, and reaches your computers over Tailscale. Armadai for iPhone has not shipped.

What Armadai has that AgentGrid's site does not list

• Several project folders on one canvas. Workspaces for different folders can sit side by side on the same canvas.
• Task boards with hand-off. A board has board, backlog, table, calendar and dashboard views. You can hand a task to Claude Code or Codex; the run gets its own Git worktree, and successful work comes back as a draft pull request for you to review.
• A built-in code editor. The Full IDE tile lets you edit code and search the project beside the agents.
• Computer use for the lead agent. The master agent opens workspaces, launches and messages agents and arranges the canvas. If you enable computer use, it can also click, type and scroll in apps on your Mac.
• Environment Vault. Secret values are stored in the macOS Keychain, and environment variables can be scoped to the app, a canvas or one workspace.
• Optional terminal recording. Terminal output can be recorded locally, off by default, with a retention window and a disk limit.

What the two have in common

Both run the agent tools you have installed, under your own accounts, and neither plan includes a Claude or Codex subscription. Both require an account. Both put terminals, a browser and notes beside the agents, give you a lead agent that starts and directs other agents from one conversation, isolate parallel work in Git worktrees, show which agents are working or waiting, and save your layout so you can come back to it. Both can have Claude Code and Codex work on the same goal: AgentGrid through mixed teams of workers, and Armadai through Coordinate, which adds a message budget and lets you review each message before it is delivered. Both can keep agents running after the main window is gone.

How the prices compare

AgentGrid's Free plan costs nothing and may be enough if you keep to three projects and run agents locally. AgentGrid Pro is $20 a month, or $192 a year billed annually, at a price its pricing page labels early-adopter, against a regular value of $50 a month or $600 a year. AgentGrid's site does not say how long the early-adopter price lasts.

Armadai has one paid plan in two billing periods: $20 a month, or $120 a year, which works out to $10 a month. Every tool is included on either plan, with unlimited workspaces and agents. The 7-day free trial requires a payment method, and annual payments carry a 30-day money-back guarantee. Current plans are under pricing.

Which one to choose

Choose AgentGrid if you use Windows or Linux, if you want a free plan, if you drive Cursor, OpenCode, Grok or another agent besides Claude Code and Codex, if you want teams of workers on different models with usage tracking across providers, or if you want to try a phone app now.

Choose Armadai if you run agents in several projects at once and want them on one canvas, if you plan work on a board and hand tasks to agents from it, if you want a code editor on the canvas, or if you want a native Mac app. If neither fits, the Maestri comparison and Conductor alternatives cover other apps for running coding agents.