LanesLanes

One workspace. Many agents. Your parallel AI advantage.

Lanes Link Is Now Available in the Desktop App, No Terminal Required
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
Lanes Link Integrations
Current Repository
my-project
Current Branch
main
ChangesHistory
Mauth.tssrc/lib+34-2
Asession.tssrc/lib+18-0
Mroutes.tssrc/api+6-2
src/lib/auth.ts+34-2
8import { sign, verify } from './jwt'
9import { createSession } from './session'
10
-export function login(user: User) {
- return { ok: true }
11+export async function login(user: User) {
12+ const token = await sign({ sub: user.id }, SECRET)
13+ const session = await createSession(user.id, token)
14+ return { token, expiresAt: session.expiresAt }
15}
16
17+export async function refresh(token: string) {
18+ const claims = await verify(token, SECRET)
19+ return login(await getUser(claims.sub))
20+}
dev/lanes/my-projectOpen in
.claude
.git
node_modules
public
src
tests
.gitignore
package.json
README.md
server.ts
tsconfig.json
package.jsonView Mode
1{
2 "name": "my-project",
3 "version": "1.4.0",
4 "type": "module",
5 "scripts": {
6 "dev": "node server.ts",
7 "test": "vitest run"
8 },
9 "dependencies": {
10 "jsonwebtoken": "^9.0.2"
11 }
12}
Runninghttp://127.0.0.1:8787/mcpProfile: work

One endpoint for every account your agents are allowed to reach.

Gmail
Connected
Calendar
Connected
Drive
Connected
Docs
Connected
GitHub
Connected
Linear
Connected
Slack
Connected
Notion
Connected
iCloud
Connected
Reddit
Connect
Discord
Connect
bunq
Connect

Get more from your coding agents.

Stop juggling terminal tabs

Every session in one place, with its status, mode, worktree and token usage.

Run ten agents, not one

Each one in its own git worktree, running at the same time. No conflicts, no work crossing branches.

Keep your own CLIs, keys and machine

The official Claude Code, Codex or Gemini CLI in a real PTY, on your auth, never an API meter.

Read the billing breakdown

Let agents run the workspace

Loops pick the next task and a verifier judges it. Over MCP, one agent dispatches three more.

Explore loop engineering

The full coding loop, in one place.

Give every agent your apps, memory and skills.

Lanes Link is one endpoint you run: GitHub, Linear, Gmail, memory and skills, behind one set of permissions.

Explore Lanes Link

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.

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.

Explore loop engineeringBuild your first loop

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.

Set up a providerRun models locally

Run the heavy work on remote GPUs.

Lanes Compute runs it on hardware you rent by the second: one A100, or a multi-node H100 cluster.

See Lanes Compute

Not just terminals. A full development environment.

Project workspace

Drag sessions through Planning, Implementation, Review and Done. Switch projects with tabs.

Live terminals

One real PTY per session for any AI CLI, with live status and plan or implement mode.

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 leaving Lanes. Inline diffs, and a history you can undo from.

Code editor built in

Monaco, the editor behind VS Code, with a file tree, syntax highlighting and markdown preview.

SQLite browser

Finds your project's databases by itself. Run read-only SQL to check what the agent actually wrote.

Local and hosted models

Point a session at Ollama on your machine, or any provider you host or buy. 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 on Cmd+Alt+1-9, injected into the agent session or run as a shell command.

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.

Full security overview

Local-first execution

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

Lanes Desktop 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.

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.

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.

claude mcp add --transport sse lanes-desktop \
  http://localhost:5353/sse
Read the MCP docs

Frequently asked questions.

No. Lanes uses your existing CLI auth, so your Claude Code, Codex, or other CLI keeps the same login and API key. We don't proxy requests through our servers, you keep paying your own provider directly.

No. Lanes is a parallel workspace around the CLIs you already use. Claude Code stays Claude Code. Codex stays Codex. Lanes just lets you run several of them side by side without conflicts or context switching.

Yes, with loops. A loop drives the Lanes MCP tools to start sessions, check results, and pick up the next issue, and a verifier decides what ships. Hard stops on tries, budget, and wall clock end it safely. See Building loops to set one up.

Locally. Everything runs on your machine: terminals, sessions, files, and CLI output. Lanes doesn't store your code, prompts, or terminal history on our servers.

Claude Code and Codex today, with more agentic CLIs added over time. If a CLI runs in a terminal, it can run in a Lane.

Yes. A provider profile points a session at Ollama or LM Studio on your machine, a vLLM box on your network, or a provider you buy by the token. Lanes sets the base URL and credential as process environment when the session starts, so requests go straight from the CLI to your provider. See Gateway and Local LLMs.

GitHub and Linear today, with OAuth import and write-back built in. Want another tracker? Request it in our Discord.

macOS Ventura or later, native on Apple Silicon and Intel. Other platforms are on the roadmap.

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

From the blog

View all

Lanes Link Is Now Available in the Desktop App, No Terminal Required

From v0.47.0, Lanes has a Lanes Link page. It installs the CLI, holds the profile and target, connects your accounts, starts the endpoint, and registers it with Claude Code or Codex. The CLI still owns consent and the token, and the app never reads either.

Lanes Link: One Endpoint for Everything an Agent Needs

Every agent wants its own copy of your accounts. Lanes Link is one MCP endpoint you run yourself: it holds them once, decides what each agent may do, and records every call. Open source, Apache-2.0, on your machine or your own cloud.

What's New in v0.46: Gateways, Local Models, and a Rebuilt Harness Page

Point a session at a model provider you host or buy: a local Ollama, a vLLM box on your network, OpenRouter, or z.ai. Lanes now installs and manages Ollama for you, and bakes a context window that survives a real agent prompt. A new Harness page detects, installs, and updates the CLIs Lanes drives, inside a settings area rebuilt around agentic coding.