Stop Watching Your AI
Relearn the Same Task
Record once. Replay forever. 85% fewer tokens.
Works with Claude Code, Cursor, Codex, Windsurf, or any AI terminal — no setup, no API keys, your existing Chrome sessions.
One click install · Auto-updates · Or grab the latest zip for dev builds
Works with these AI coding agents
Your AI is wasting tokens doing the same thing every day.
Every task, it re-explores from scratch. AgentLimb fixes that.
AI has no memory. Give it one.
Every time your AI posts on Reddit, it rediscovers where the button is. Every time it fills a form, it re-reads the page from scratch. You're paying tokens for work it already did yesterday.
AgentLimb records that route as a "muscle." Next run: one command, zero exploration, the same result — without the re-learning tax.
One exploration per platform. Saved permanently. The more you reuse, the more you save.
Three layers. All localhost. Zero cloud.
16 standardized tools, CDP-native. Any AI terminal that can send HTTP can drive your browser.
People who do the same thing every day.
You're not lazy. You just know that repetition is a waste of your time — and your tokens.
Every other approach has a real cost.
Browser automation has been solved — badly — four different ways. Here's the honest breakdown.
| Browser Use / Playwright | BrowseAI / Browserbase | Codex / Claude Computer Use | AgentLimb | |
|---|---|---|---|---|
| Setup | Write scripts, manage dependencies, handle headless mode | SaaS config, per-workflow setup | Mac only (needs desktop env), sandbox required | Copy one prompt. Done. |
| Element targeting | CSS/XPath — you write and maintain them | Visual AI detection — unstable on updates | Screenshot coordinates — ±1px can miss the target | CDP reads the live DOM — semantic, precise |
| Token cost per action | None (pure script) | Cloud fees + AI tokens | 1,000–3,000 tokens/screenshot × every step | ~300 tokens/step. Hot start: 85.7% less. |
| Repeat task cost | Fixed (script re-runs) | Linear — billed per execution | Linear — re-explores every time, no memory | Decreasing — muscle memory compounds |
| Login sessions | Extra cookie/session setup | Cloud — can't use your local sessions | OS-level, unaware of browser state | Your real Chrome — already logged in |
| When the site updates | Selectors break — rewrite scripts | Visual model may degrade silently | Screenshot inference adjusts, but expensively | AI detects mismatch, finds new selector, self-heals muscle |
| Data privacy | Local ✅ | Through third-party servers ❌ | Local ✅ | 100% local — 127.0.0.1 only ✅ |
| AI terminal choice | Any (pure script) | Varies by platform | Bundled with Codex / Claude | Any AI that speaks HTTP |
| Shared knowledge | Script = locked to one AI | Workflow = locked to platform | No persistent memory | Muscle files = cross-AI, transferable, permanent |
Muscle memory is cross-AI: knowledge explored by Claude Code today is reused by Codex tomorrow — same JSON files, same Desktop folder.
Up and running in under a minute.
No account. No API key. No cloud. Just a Chrome extension and one prompt.
Install the Extension
One click from the Chrome Web Store — auto-updates. Or load the latest zip manually from GitHub Releases for dev builds.
Start the Bridge
Run the bundled Node.js bridge once (macOS users: scripts/install.sh registers a LaunchAgent so it auto-starts on login).
Copy the Prompt & Paste
Open the side panel, click "Copy Onboard Prompt", paste to Claude Code / Cursor / Codex / any AI terminal. It auto-connects and starts working.