Lanes does not replace Claude Code. It runs it. If you already work in Claude Code and want to run more than one session at a time with real oversight, Lanes is the workspace built for that.
The split is simple: Claude Code is the agent, Lanes is the workspace around it. Nothing about the CLI changes, and everything about running several of them does.
What Lanes adds to Claude Code
- Parallel sessions. One per issue, each the official
claudeCLI in a real terminal. A session is a PTY attached to its issue, so the task, transcript, and results stay together; close the panel and it keeps running. One issue can also hold several named sessions (plan, implement, review) that share the issue's worktree but keep separate transcripts, status, and metrics. - Isolation. A git worktree and branch per session, created and cleaned up automatically. Worktrees live under
.worktrees/off your base branch; dirty ones trigger a warning before removal, so uncommitted work is never dropped silently. - A board. Plan, run, review, and ship, with live status on every session and a bell when one needs you. Cards move through Planning, Implementation, Review, and Done, and the Meta tab keeps tokens, cost, and runtime per session.
- Your plan, not a meter. Lanes runs the real CLI, so sessions draw from your existing subscription. That distinction got sharper in June 2026, when Anthropic moved Agent SDK usage onto separate credits while interactive CLI use stayed on plan limits. Anthropic just split the bill covers what changed and who it affects.
- Orchestration. A Lanes Desktop MCP server lets an agent manage the board and spawn more sessions. It is a research preview with 30 tools over SSE, covering issues, sessions, and the GitHub and Linear bridges.
- Somewhere else to get tokens, when you want it. A Gateway profile points a session at a provider you host or buy, and Lanes derives the environment variable names Claude Code expects, so the same profile works whichever CLI the session picked. It is applied as real process environment at launch, so the request goes straight from the CLI to your provider and the credential never reaches scrollback or
ps. Local LLMs is the managed version of that: Ollama installed, started, and pulled from one page, with the context window sized so a real agent prompt is not silently truncated. - The binary itself, kept current. The Harness page shows whether
claudeis on your PATH, the version it reports, and which package manager installed it, then updates through that manager rather than assuming npm.
Two Claude-specific details are worth knowing. Stopped sessions restart with --resume, so pausing an issue does not lose the conversation. And settings are per session: pick the model and reasoning effort in the launch flyout, and run one issue on a heavy model while five light ones run beside it.
When you need it
One Claude Code session is great. The moment you want several working at once without tripping over each other, you need a workspace around them. That is what Lanes is.
Claude Code stays exactly as it is. Same binary, same auth, same behavior. Lanes is the surface that lets you run a fleet of them.
In practice the shift is workflow, not tooling. Break work into issues, run several sessions side by side, review each branch's diff before merging, and keep the billing math in view as the fleet grows.
Further reading
- Quick start: install Lanes and run your first session in about two minutes.
- Working with sessions: lifecycle, statuses, sibling sessions, and per-session settings.
- Worktree management: isolation, dirty-state warnings, and cleanup.
- Manage multiple Claude Code sessions: the day-to-day workflow.
- Point sessions at any model provider: profiles, auth modes, and what you give up.
- Anthropic just split the bill: the June 2026 billing change explained.
Get Lanes Desktop or read the quick start.