Right now, when your OpenClaw agent visits a website, it's using a full Chrome browser under the hood. That works, but it's slow, uses a ton of memory, and can't log into anything without you manually setting up sessions.
Want your agent to check your X mentions? Read your GitHub notifications? Monitor a LinkedIn thread? It can't, because it's not logged in, and logging in with a headless browser triggers bot detection on every major platform.
That's Plasmate.
Plasmate compiles HTML into structured data without rendering pixels. No GPU, no layout engine, no wasted work.
Log into X, GitHub, or any site yourself. Hand the cookies to your agent. It browses as that account.
Pages become structured "Semantic Object Models" instead of raw HTML. Your agent understands pages, not pixels.
~15MB binary. No Chrome download. Runs on a Raspberry Pi. Uses a fraction of the memory.
Three pieces, all free and open source. Setup takes about five minutes.
Install with one command:
This puts the plasmate binary on your machine. It's a single file, no dependencies.
Install the Plasmate skill so your agent knows how to use it:
Once installed, your agent automatically uses Plasmate for web browsing when it makes sense. No configuration needed.
This is what lets you hand login cookies to your agent. Install it once:
You'll see a small Plasmate icon appear in your Chrome toolbar.
Chrome Web Store listing is coming soon. For now, the manual install takes about 30 seconds.
Let's walk through the most common use case: letting your OpenClaw agent browse X/Twitter as a logged-in account.
You log into X in Chrome like normal. The extension grabs the auth cookies. Plasmate stores them encrypted on your machine. Now your agent can browse X as that account.
Open a terminal and run: plasmate auth serve
This starts a local server that the extension talks to. Leave it running.
Go to x.com and log into the account you want your agent to use. This could be your personal account, a brand account, or an account you created specifically for your agent.
A popup shows all cookies for x.com. The important ones (ct0 and auth_token) are already highlighted and selected.
One click. The cookies are encrypted and stored locally. The extension confirms with a green message. You're done.
That's it. Your agent can now browse X as that account. It can read timelines, check mentions, view DMs (if the cookies include that access), and interact with posts.
Cookies expire eventually. Plasmate tracks expiry dates and warns you when cookies are getting stale. When that happens, just repeat steps 2-4. Takes about 10 seconds.
The same process works for any website. The extension auto-selects the right cookies for popular platforms:
Read timelines, check mentions, monitor keywords, view analytics.
Review PRs, read private repos, check notifications, browse issues.
Read messages, check connection requests, monitor company pages.
View feeds, check DMs, monitor engagement on posts.
For any other site, just navigate there in Chrome, click the extension, select the cookies that look like session or auth tokens, and push. If you're not sure which cookies matter, select all of them.
Check which sites your agent can access:
Remove access for a site:
Once Plasmate is installed and you've pushed cookies for the sites you care about, your OpenClaw agent can do things like:
The agent uses Plasmate automatically through the skill. You don't need to tell it "use Plasmate" or configure anything per-request. It just works.
Launches Chrome, renders pixels, takes screenshots, uses OCR to read them. Slow, heavy, can't log in.
Compiles HTML directly into structured data. Understands forms, buttons, links, and content natively. Fast, light, authenticated.
Questions? Check docs.plasmate.app for the full documentation, or ask in the OpenClaw Discord community.