A real Linux sandbox, any model you choose. No walled garden.
TL;DR
Most coding agents lock you into one vendor's model. rigless doesn't — bring your own OpenAI, Anthropic, Google, OpenRouter, Moonshot, or Zhipu key, or start on a bundled DeepSeek tier.
Every session runs in a real, isolated Ubuntu Linux VM, not a simulated sandbox.
A real git workflow: clone, branch, and open a pull request without leaving the browser.
Works from any device with a browser. No local install.
Almost every coding agent shipped in the last two years makes the same bet: pick one model, wrap a slick interface around it, and hope nobody asks what's running underneath. Devin bets on its own orchestration. Cursor and Copilot bet on whichever model they've licensed this quarter. Claude Code bets on Claude. Lovable, Bolt.new, and v0 bet on speed to a demo.
We think that's the wrong bet. rigless is built differently on two axes: the model is your choice, and the environment is a real computer.
Every coding tool eventually hits the same wall — the model changes. A better one ships, your favorite one gets deprecated, your company standardizes on a different provider. If your agent is hard-wired to one vendor, that wall is a wall for you too.
rigless doesn't hard-wire anything. Bring your own key for OpenAI, Anthropic, Google, OpenRouter, Moonshot, or Zhipu, or point it at any OpenAI-compatible endpoint you run yourself. We bundle a DeepSeek tier so you can start in minutes — that's onboarding fuel, not the pitch. The agent loop, the sandbox, and the git workflow stay identical no matter which model is reasoning. A better model ships next month? Point rigless at it with your own key and keep everything else.
That's not "our model is smarter than theirs." We're not making that claim. The value isn't which model you use — it's not being stuck with our choice of one.
A lot of agent products are chat interfaces with a diff viewer bolted on. The "execution" is simulated, sandboxed too tightly to install a real dependency, or running on infrastructure you can't see.
Every rigless session gets a real Ubuntu Linux sandbox. Install anything, run a database, start a dev server. It's isolated with microVM-level hardware virtualization — a real filesystem, a real terminal, a real git remote. When the agent says it ran your test suite, it ran your test suite. Open the terminal and watch it happen.
And it's a real git workflow, not an export button: clone a private repo, work on a branch, open a pull request for human review, all without leaving the page.
The sandbox lives in the cloud, not your laptop. rigless works from a phone, a tablet, a Chromebook, or a machine you don't own. No CLI to install, no local environment to configure, no admin rights required. If it has a browser, you can ship from it.
Other agents | rigless | |
|---|---|---|
Model | Fixed to one vendor | Any — BYO key or bundled starter |
Environment | Varies: local, opaque cloud, or simulated | Real isolated Linux VM, every session |
Access | Often one device or one IDE | Any browser, any device |
Git | Varies | Clone, branch, PR — built in |
You want one vendor to make every decision for you. A fully managed, single-model agent removes choice — sometimes that's what you want.
You're fully committed to one IDE or platform already and don't need portability.
You want the freedom to switch models without switching tools or rebuilding your workflow.
You want to see and touch the real environment the agent is working in, not trust a black box.
You work from more than one device, or don't want to manage a local toolchain.
We'll go deeper on specific comparisons in the posts that follow — Devin, Replit Agent, Cursor and Copilot Workspace, Lovable/Bolt.new/v0, Claude Code, and free aggregators like Freebuff. Start a 7-day trial, no credit card required, and bring your own keys whenever you're ready.