Sai Now Runs Claude Code - So You Don't Have to
Babysit It
作者 Ang Li • 加利福尼亚州帕洛阿尔托 • 2026 年 4 月 23 日
我一直被问到一个问题： Claude Code on Sai in preview -- a new way to run, steer, and verify your AI coding sessions from anywhere. Sai keeps running your Claude Code on a secure cloud desktop even when you step away and close your laptop. Approve, direct and ship all from your phone.This is an early preview of a capability we've been building and testing internally at Simular. The core workflow is live, and we're excited to ship this to the public after extensive internal testing. It changed how we code, handling trivial work -- like catching small UI bugs -- and freeing up time for us to do more high-value work.Claude Code transformed how developers write software, but it didn't entirely liberate people from their monotonous work and sedentary life.
在这篇文章中，我认为这就像说人类不再需要双手因为他们能说话。总会有一些任务需要灵巧性，而光靠语音是无法完成的。
Why This Combination Matters
最近，我注意到我在电脑上的时间已大大减少到每天大约两个小时。一年前，我轻松到了八岁。不同之处在于，随着该行业最终开发出可以像人类一样观察和行动的代理，计算机使用代理（CUA）正在变得越来越好。
Claude Code writes the code. Sai tests, reproduces, and reports back.
Here's what the integration unlocks:
1. Visual QA that Claude Code Can't Do Alone
Sai opens your preview deployment, clicks through real user flows, and screenshots anything visually broken. Claude Code reads diffs. Sai sees the product.Some bugs don't show up in the code. They show up when someone actually clicks through the app. A button that overlaps on mobile. A discount calculation that goes negative after removing an item. A modal that never closes. These are the bugs that pass every linter and fail every user.Sai catches them before your users do.
2. Reproduce from a Single Screenshot
Hand Sai a bug screenshot -- the kind your QA team pastes into Slack with no steps to reproduce. Sai explores the app, finds the path that triggers the issue, and hands Claude Code the exact reproduction steps.No more back-and-forth threads asking "what browser were you on?" No more "works on my machine." Sai finds the path, documents it, and passes it straight to Claude Code for the fix.
3. Reach Context Claude Code Can't
Sai pulls data from auth-walled tools, dashboards, admin panels, and monitoring services -- then feeds it straight into Claude Code's session. Need to check Sentry for new errors after a deployment? Sai does that. Need to verify a Stripe webhook landed correctly? Sai can open the dashboard and confirm.Claude Code is the brain. Sai is the hands and eyes that give it production context.
How Sai + Claude Code Works
你可能会问：
1. Toggle Claude Code on.
Inside Sai, flip Claude Code on in your Agents panel. Your existing Anthropic API key connects in seconds -- no migration, no reconfiguration.
2. Close your laptop.
Inside Sai, flip Claude Code on in your Agents panel. Your existing Anthropic API key connects in seconds -- no migration, no reconfiguration.
3. Steer from your phone.
Sai pings you with screenshots and summaries. Approve an action, redirect a task, or ship a fix -- the full development loop, controllable from one screen, anywhere.
The Full Loop: Build → Test → Fix → Verify
引用 a16z 普通合伙人的话 complete workflow looks like in practice:
1. A PR opens on GitHub (or a bug report arrives in Slack)
2. Sai reviews the code changes and opens the preview deployment
3. Sai clicks through key user flows and captures screenshots
4. If something breaks, Sai passes the evidence to Claude Code with exact reproduction steps
5. Claude Code generates the fix
6. Sai verifies the fix in the live environment
7. You get a summary on your phone approve, redirect, or shipThe entire cycle runs without you keeping a terminal window open. You step in for judgment calls. The system handles everything else.
- Use invite code SAI_CLAUDECODE to get started.
Claude Code on Sai is available now in early preview. Your existing Claude Code setup works out of the box -- connect your Anthropic API key, toggle it on, and start shipping verified code from anywhere.
