Questions

Does it work with physical devices?

No. A lane holds an iOS simulator or an Android emulator from agentlane's pool. Those are what several agents can share on one machine without stepping on each other, and what up can create, boot, set up and reset on its own. Physical phones are outside what agentlane manages.

Does agentlane send anything off my machine?

agentlane has no telemetry. It makes network requests only to download its pinned tools on first use: its Maestro build from agentlane's GitHub releases, Playwright and agent-browser from the npm registry, and Chrome for Testing and ffmpeg from Playwright's CDN. Each download is checked against a sha256 the package pins. Where it runs lists them.

Stock Maestro calls its makers' servers for analytics, an update check, error reports and a report after maestro test. agentlane's build removes all four, its build fails when any of that code is still in it, and a live check watches a whole Maestro session for connections off the machine. Your own app and its servers reach whatever they reach; when a dev server's CLI sends usage telemetry, the setup skill turns it off in the service's start command.

Does it run on Linux or an Intel Mac?

Linux on x64 and arm64 runs the web half of a project: ports, services, resources and a browser per lane. That covers cloud agents and Linux machines. iOS needs a Mac, and Android on Linux isn't supported yet. The browser builds for Intel Macs are pinned, but no Intel Mac has run agentlane, so that host is untested. Where it runs has the table.

Does it work in a monorepo?

Yes. up, doctor and restart read agentlane.json from the current folder or the nearest folder above it, so the config can sit at the repository's root or beside one app. Each service's cwd points at its package, and init writes the agents' files beside the config. When it installs agentlane, init uses the nearest package.json and the workspace's package manager, with --workspace-root at a pnpm workspace root. The build cache lives in the repository's git folder, so every worktree shares it.

Can I use it in CI, or without an agent?

Yes. Every command works from a plain shell, so a CI job can bring the web app up, run the project's tests in the lane, and release it:

npx agentlane up --platform web
npx agentlane exec -- npx playwright test
npx agentlane release

Without an agent, a lane belongs to the shell that claimed it, and it is reaped once that shell is gone. So run the three commands in one script or CI step, or set AGENTLANE_OWNER_PID to a process that lives as long as the lane should. Browsers run headless by default, and emulators always run without a window. On a slim Linux image, install Chrome's system libraries first; Troubleshooting shows how.

How many agents can run at once?

As many as the machine has room for. Ports come from blocks of 20 per service type. Devices come from a pool of 5 simulators and 5 emulators by default, inside a 20 GB memory budget, and you can raise all three in the registry's config.json. Web lanes need no device; the port blocks allow 20 lanes per service type, and you can widen them. When the pool is full, the next agent waits in a queue instead of failing. What a lane holds has the limits.

Can subagents of one session share a lane?

They shouldn't, and they don't need to. A lane belongs to a session and a folder, so subagents that each work in their own worktree each get their own lane, with its own device, ports and data.

Does agentlane change my app's code?

Only where it can't work otherwise. A server that hard-codes its port has to read it from AGENTLANE_<NAME>_PORT, and a mobile app has to take its API address at launch rather than from the build. The setup skill makes those changes, keeps your existing flows and tests passing, and lists every change in its summary.

What happens to a lane's data when it ends?

Its resources' teardowns run, so a per-lane database is dropped. The lane's evidence folder, with its screenshots, videos, traces and logs, stays for seven days after the lane ends.

What does it cost?

agentlane is free and MIT licensed. What it costs your machine is disk for the pool's simulators and emulators, which prune --yes gives back, and the memory of the devices your lanes hold. For your agents' token budgets, look returns the screen as a few lines of text, which costs far fewer tokens than a screenshot, and the warm driver keeps each call short. What makes it fast has more.

Can I run 0.3.0 beside 0.2.x?

Yes, on the same Mac, so projects can move one at a time. Both versions share one registry and see each other's lanes. The changelog has the rules and the steps to move a project.