Jev
Haiku
Agent
Free
$0/forever
Open source. Bring your own key.
- Full read-only enforcement
- Bring your own model key (BYOK)
- Open source, inspect every line
OnlyRead, the read-only mode for AI browser agents.
OnlyRead
is the read-only mode for AI agents in your browser. Your agent can explore your sites. It just can’t act on them.
Go on. Try to like it. Tap the switch to turn it off.
Please don’t change anything 🙏 Just a request
OnlyRead makes it
a boundary.
A little thought before every action.
1,000+real tasks, tested.
Validated against a published university benchmark of AI agents working on the web’s busiest sites. OnlyRead holds up wherever people actually browse.
Add it to Chrome
One click from the Web Store. It lives in the browser your agent already uses.
Pick your apps
Choose the sites to protect. It guards them whenever an agent is working.
Let it look around
Your agent reads everything it needs. Anything that would change something stops at the glass.
Pick your access level
Pro
Free while in early access. Paid Pro will be $5 a month.$5/month
Fully managed. Nothing to set up, nothing left out.
Everything in Free, plus:
- No AI API key to set up or top up
- Central control: your guarded sites in every browser, from one account
- Set up any AI agent’s browser from one link
- The best judge models, updated for you
No credit card required
Enterprise
Let’s talk
For teams putting agents to work at scale.
Everything in Pro, plus:
- One place to see every browser agent’s activity across your team
- Integrations with the tools you already use
- Dedicated support
- Custom app testing and verification
Good questions.
Straight answers.
Is this another AI agent?
No. OnlyRead is a free, open-source Chrome extension that gives AI browser agents read-only access to your web apps: they can read and search, but cannot send, post, delete or buy by mistake. You bring the agent; OnlyRead limits what can change in the apps you guard.
What does “read-only” actually mean?
Requests classified as changes are blocked; reads continue. Some apps use the same endpoint for reads and writes, so app-specific rules and request classification matter. It is a strong boundary, not a guarantee against every possible change.
Can the AI judge get it wrong?
Rarely, but yes. Most requests are settled before a model is asked: by rules, per-app decoders and verdicts reviewed by hand. For the rest, the judge is tested on real requests recorded from the apps we support, and anything it isn’t sure about is refused. Like any AI model, it can still misjudge a request, so treat OnlyRead as a strong safety net, not a guarantee, and keep an eye on what your agents do in the accounts that matter.
Does my data leave the browser?
Your settings and decision log stay in Chrome storage. Most requests are decided locally. The few that aren’t go to a model for a verdict: on Free, to your own provider with your key; on Pro, to OnlyRead’s service, which never stores the request content.
Where do I get it?
From the Chrome Web Store, in one click. It runs in Google Chrome and Chromium browsers such as Perplexity Comet, with any agent that drives them: Claude in Chrome, ChatGPT in Chrome and Codex, Gemini in Chrome, Claude Cowork and others. Every site starts off; you switch on the apps your agent may use, and it guards them whenever an agent is working. The full walkthrough, with a test you can run, is in how to make an AI browser agent read-only.
Give your agent
the right kind of access.
All that curiosity. A little more control.
