Use cases/Lanes Desktop/Run agent fleets on your Claude subscription

Run agent fleets on your Claude subscription

Lanes runs the official CLI in a real terminal, so parallel sessions draw from the plan you already pay for, not a separate meter.

Many agent tools drive Claude through the Agent SDK or an API key, which meters usage separately from your plan. Run several at once and the cost climbs fast. Lanes takes a different path.

How Lanes does it

  • The real command. Lanes opens a real terminal and runs the official claude command, exactly as you would by hand. A session is a real pseudo-terminal booted from your login shell with your dotfiles, not an SDK call dressed up as one.
  • Interactive, not programmatic. From the provider's side, a Lanes session is interactive CLI use, so it draws from your existing subscription. Anthropic's June 2026 change moved Agent SDK and claude -p usage onto a separate monthly credit; interactive Claude Code in the terminal stayed on plan limits, and that is the surface every Lanes session runs on.
  • No proxy. Lanes does not route your requests through its servers. You keep paying your provider directly.
  • Same for Codex. The official CLI, your login, your plan.

The same holds when sessions stop and restart. Lanes relaunches Claude Code with --resume, so the conversation continues under the same login instead of starting a metered job somewhere else. And because Lanes launches the CLI from your shell, your existing auth just works; there is no API key to create and nothing new to pay for on the provider side.

You can also see what you are using. Each issue's Meta tab tracks tokens, cost, and runtime per session, and the status bar totals tokens and estimated cost across everything running. Per-repo settings control the flags and environment variables sessions launch with, and Lanes records what each session actually ran with.

What Lanes adds on top of the CLI is the workspace: parallel sessions, worktrees, and the board. Lanes with Claude Code covers that side.

Why it matters

You can run a fleet of parallel agents without a surprise metered bill. Your plan keeps doing what it did before.

This is an architecture property, not a pricing trick. Tools built on the Agent SDK got repriced in June 2026 because of what they are built on; tools that run the interactive CLI did not. If the line moves again, which side you are on is decided by the runtime, and the runtime here is the official CLI in a terminal.

Set it up

Install Lanes and open it:

Bash
brew install --cask lanes-sh/lanes/lanes && open -a Lanes

Then:

  1. Check that claude works in your terminal and is logged in to your plan. Lanes runs the same binary with the same auth.
  2. Add a project and create an issue with Cmd+N.
  3. Start a session with Plan or Implement and watch the Meta tab count tokens against your plan, not a separate meter.

The quick start covers the first run in about two minutes.

Further reading

Read the full breakdown or get Lanes Desktop.