One workspace. Many agents. Your unfair advantage in parallel AI coding.

Lanes
my-project
Backlog2
#742Add dark mode
#743Add user authentication
Planning & Implementation1
#737Fix server crash on corrupted db.json
Done2
#739Set up CLI pipeline
#734Set up git and repo
SessionHistoryChanges
Fix server crash on corrupted db.json
Claude Code v2.1.92
Opus (with 1M context) · Lanes Team
Tips for getting started
Run /init to create a ...
Recent activity
No recent activity
Ready to get started?
#743 Add user authentication
Claude Code v2.1.92
Opus (with 1M context) · Lanes Team
Tips for getting started
Run /init to create a ...
Recent activity
No recent activity
Implementation completed
#737 Fix server crash on corrupted db.json
3 files changedTests passing

Get more from your coding agents.

Stop juggling terminal tabs

Multiple terminals, multiple agents, and the constant question: which one is on which branch? Lanes shows every session in one place. Status, mode, worktree, token usage. Nothing hidden behind a tab.

Run real CLIs, switch anytime

Lanes orchestrates the official CLI for each agent in a real PTY: your shell, your dotfiles, your interactive plan, never an SDK or API meter. Run Claude Code today, Codex tomorrow, Gemini CLI next week, even within the same project.

Read the billing breakdown

Isolate every session

Assign different tasks to different agents, each running in its own worktree with custom setup and teardown scripts. No conflicts, no context switching, no work crossing between branches.

Run agents unattended

Define a loop once: a skill for the work, a verifier the model cannot talk its way past, hard stops for tries, budget, and time. Lanes runs it while you do something else. The loop, not the prompt, becomes the unit of work.

Read the essay

The full coding loop, in one place.

Integrations

Pull work from GitHub, Linear, or any MCP chat.

Connect your issue tracker once. Lanes imports tickets, spins up an isolated worktree per task, and posts results back as PRs or comments when the agent is done.

Isolated worktrees

Run parallel agent sessions on auto-created git worktrees.

One issue, one branch, one isolated workspace per agent. No conflicts. No "wait, which folder is this?" Auto-cleanup when the session is done.

Loop engineering

Run loops, stop prompting.

A loop prompts the agents for you: pick the next task, run a session, let a verifier judge the result, repeat. Hard stops end it on tries, budget, or wall clock. The board is its state, worktrees its isolation, the MCP server its hands.

Model providers

Point any session at the model provider you choose.

Run against Ollama on your own machine, a box on your network, or a provider you buy. Lanes sets up the local ones for you, and requests go straight from the agent to your provider, never through us.

Project workspace

Drag sessions through Planning, Implementation, Review, and Done. Switch between projects with tabs, multi-select, and context menus.

Live terminals

Real PTY sessions for Claude Code, Codex, or any AI CLI, one per session. Real-time status, plan or implement mode, and resume across restarts.

Resume sessions

Restart without losing context. Lanes resumes Claude Code sessions with full conversation history automatically.

Git client, not just a diff

Review, commit, publish, and switch branches without opening GitHub Desktop. Repository and branch header, inline diffs, and a history you can undo the last commit from.

Code editor built in

Monaco, the editor behind VS Code, with a project file tree, syntax highlighting, save on Cmd+S, and side by side markdown preview.

SQLite browser

Finds the databases in your project by itself. Browse tables and views and run read-only SQL to check what the agent actually wrote.

Local and hosted models

Run agents against Ollama on your own machine, or point a session at any provider you host or buy. Managed in settings, with nothing routed through us.

Dependencies

Link issues as dependencies with automatic cycle detection. Blocked issues unlock when prerequisites reach Done.

Quick commands

Preset and custom commands triggered with Cmd+Alt+1-9. Two types: CLI (injected into the AI coding session) and terminal (run as shell command).

Lanes MCP server

Tell one agent to dispatch three more

From Claude Code, Codex, or any MCP enabled chat, start sessions, watch their terminals live, and triage the board without leaving the conversation. The same tools power loops, no human in between.

claude mcp add --transport sse lanes-local \
  http://localhost:5353/sse
Lanes Forms

Give your agents a forms backend, the easy way.

Create secure, agent friendly forms in seconds. No backend, no setup. Capture, protect, and route submissions anywhere.

Get started in seconds.

Mac app today. Native on Apple Silicon and Intel, requires macOS Ventura or later. Join our Discord community for updates and feedback.

Security

Built to be trusted.

Your code, prompts and terminal output stay on your machine. When you opt into remote workspaces, data lives in the EU with a public list of every sub-processor we use.

Local-first execution

Source code, terminal output, prompts and agent responses never leave your machine.

Local MCP server

The Lanes MCP server runs locally on your machine.

Encrypted in transit and at rest

TLS 1.2+ and Google Cloud platform-managed encryption.

Frequently asked questions.

The workspace for parallel coding agents.

One workspace to plan, run, loop, and ship with AI agents.