What is Lot
Lot is a cloud platform for running coding agents in fully replicated dev environments. Fire off tasks in parallel, let agents work while you're away, and come back to verified, reviewable results — from any device.
The problem with local agents
Tools like Claude Code run directly on your machine. They're powerful when you're sitting at your laptop, but they hit a wall fast:
- Sessions die when you close the lid — the agent stops when your laptop sleeps. Long tasks require keeping your machine running and unattended.
- Parallel agents are limited — git worktrees solve the dirty-repo problem, but port conflicts remain. Two agents both trying to run a dev server on port 3000 means one of them can't do e2e verification. There's no clean way to run multiple agents with auto-verification locally today.
- You can't steer it when you're away — even if your machine is caffeinated and the agent is running, you can't check on it, unblock it, or redirect it from your phone. If it gets stuck on a question, it just waits until you're back.
The problem with cloud coding agents
Cloud agents like Codex and Devin solve the "always running" problem, but they don't replicate your dev environment. They load the codebase, make changes, and produce a diff. There's no running app, no dev server, no runtime.
- You can't verify remotely — the agent produces code, but unless you pull the changes and run the app yourself, you're reviewing a diff on faith.
- The agent has no feedback loop — without a running app, the agent can't tell if the UI renders correctly, if the page loads, or if it broke something. It writes code in the dark.
- No runtime means no verification — auto-verification needs a live app to test against. Without a dev server, there's nothing to launch a browser on. This makes long, multi-step tasks unreliable — the agent drifts at step 3 and you don't know until you review the final diff.
What Lot does differently
Every Lot session runs in a full cloud sandbox — your repo cloned, dependencies installed, dev server running. The agent doesn't just edit files; it works inside a live environment where changes can be seen, tested, and verified.
Long tasks that actually work
With auto-verification enabled, a separate verification agent launches a real browser after each change. It checks that the app still works, takes screenshots, and reports back to the coding agent. The agent self-corrects along the way instead of drifting for 10 steps and handing you a broken diff.
This makes complex, multi-step tasks practical as async work. Fire off a task, close your laptop, and come back to verified, working code.
Parallel agents without the pain
Each session is an independent sandbox with its own resources, ports, and git state. No conflicts:
- "Add dark mode" in one session
- "Fix the login bug" in another
- "Write tests for the API" in a third
They all run simultaneously, each verifying their own work.
Access from anywhere
Everything runs in the cloud, so you can work from any device with a browser — your phone on the bus, a tablet on the couch, or a borrowed laptop.
- Start a new task — open the workspace, create a session, and send a message. No local setup, no git clone, no dependency install.
- Check on a running task — see the live preview, read the real-time diff, and watch the agent work. If it's going well, close the tab and check back later.
- Continue an existing task — review what the agent did, leave follow-up instructions, answer its questions, or redirect its approach. Use plan mode to approve the next steps before any code is written.
- Ship when ready — create a PR directly from the workspace. No local checkout needed.
How it works
- Connect your GitHub repo — Lot links to your repository via a GitHub App.
- Automated setup — an AI setup agent installs runtimes, services, and dependencies, then takes a snapshot. Future sessions boot from this snapshot in seconds.
- Start sessions — each session gets a sandbox with your code, a running dev server, a live preview, and a Claude agent. Send a task and let it work.
- Review and ship — check diffs, preview the app, and create a PR when you're satisfied — all from the browser.
What you get
| Feature | Description |
|---|---|
| Cloud sandboxes | Isolated Linux environments with 2-4 vCPUs and 4-8 GB RAM |
| Auto-verification | Browser agent tests the app after each change, catches errors, takes screenshots |
| Live preview | See your app running as the agent makes changes, from any device |
| Parallel sessions | Run multiple agents on different tasks simultaneously |
| Real-time diffs | Syntax-highlighted file diffs updated as the agent codes |
| PR creation | Create and update pull requests directly from the workspace |
| Terminal access | Full terminal into the sandbox for manual commands |
| Environment variables | Securely store and inject secrets into every session |