Safeguard Browser Extension Guide: Install, Ask and Hand Over a Tab

The Safeguard browser extension puts Safeguard Search in a side panel next to the page you are on. Install it from the Chrome Web Store, Firefox Add-ons or the Safeguard iOS app, sign in, and ask about your findings, projects and compliance posture. In Chrome you can also hand Search a tab and approve its risky steps.

Safeguard is one platform on every surface, and the browser is one of them. You do not lose the AI you already pay for when you use it. Bring the Claude, ChatGPT or Codex plan you already pay for and use it inside the Safeguard CLI. Safeguard adds no extra charge for that usage; your plan's own usage limits and terms still apply. You can also connect Claude or ChatGPT to the Safeguard MCP server and use Safeguard from the chat you already use.

What you will have at the end of this guide:

1. The Safeguard extension installed and signed in on the browser you use every day.
2. A short list of side panel questions you can ask each morning about your own findings.
3. Browser control set up in Chrome with the approval mode that fits the task.
4. The Safeguard CLI on your machine for the work a browser cannot do: editing code, running tests and scanning folders, with your approval.
5. A troubleshooting checklist for the problems people hit in the first week.

Install and sign in, step by step

Pick the right listing for your browser

There is one extension, version 1.0.4 on the download page, published by Safeguard.sh Inc. It ships through three stores, and what it can do depends a little on the browser. This table uses only the download page, the store listings and the Safeguard docs, all read on 21 October 2026.

The docs list the minimum versions: Chrome 114 or later, Edge 114 or later and Firefox 113 or later. If the panel is blank, check your browser version first.

The extension needs a Safeguard account. If you do not have one yet, the download page offers a cloud sandbox tenant with a seeded demo repo, free for seven days with no card, which is a good place to try everything in this guide on test data.

Chrome, Brave, Arc, Vivaldi, Opera and Edge

1. Open the Safeguard listing on the Chrome Web Store (linked from safeguard.sh/download).
2. Select Add to Chrome. When the browser shows the permission prompt, read it, then select Add extension.
3. Open the Extensions menu (the puzzle piece icon), find Safeguard and select Pin so the icon stays on the toolbar.
4. Click the Safeguard icon. The side panel opens and shows the sign in screen.
5. Sign in with your Safeguard account. When the search page replaces the sign in screen, you are ready.
6. Ask one question, such as "how many critical vulns do we have", and check that an answer renders in the panel.
7. Close the panel and open it again. You should still be signed in.

The permission prompt in step 2 mentions debugging. That is expected. The store listing explains that the extension uses Chrome's debugger permission to send genuine input, so file pickers, native form controls and drag and drop work when you later hand it a tab. Chrome does not allow that permission to be optional, so the warning appears on every install, whether or not you ever turn browser control on.

Firefox

1. Open the Safeguard listing on Firefox Add-ons and select Add to Firefox.
2. Open the sidebar. If you closed it, use View, Sidebar, Safeguard, or press Ctrl+B and pick Safeguard from the sidebar switcher.
3. Sign in. The Firefox listing says you approve the browser from your Safeguard account, much like the CLI signs in, so you never type a password into the extension.
4. Run a test question and confirm the answer appears.

Safari on iPhone and iPad

1. Install the Safeguard.sh app from the App Store. The Safari extension is included in it.
2. Open Settings, Apps, Safari, Extensions and turn Safeguard on. On releases before iOS 18 the path is Settings, Safari, Extensions.
3. Open Safari and start the extension from there.

Safari on Mac comes with the Mac app from the same App Store listing, and the download page marks it as in Apple review. Until that clears, use Chrome or Firefox on your Mac, or the desktop app.

Ask from the side panel every day

The side panel answers from your real Safeguard account, not from general knowledge. The store listings describe four kinds of question it handles well: open findings and their severity, which projects are affected, what a CVE means for the version you ship, and where you stand against the compliance frameworks you track.

It uses the same conversation as the Safeguard web app, so a thread you start in the panel is still there later in the browser, the desktop app and the phone app. You can attach a file or a screenshot when a question needs one.

A five minute morning routine

Open the panel on whatever page you start your day on and work through these:

1. "How many critical vulns do we have, and how many are new since yesterday?"
2. "Which projects are affected by the critical findings, and which have a fixed version available?"
3. "Summarise the open findings on payments-api by severity."
4. "Which controls are failing in the frameworks we track?"
5. "Draft a short note for the team channel listing the top three things to fix today."

The last answer is a draft. Copy it, read it and send it yourself.

Ask about the page you are on

Some of the most useful questions are about a page you are already reading. In Firefox, the listing describes attaching the current tab and asking about it; the sidebar reads the page only when you ask. Good examples:

• On a vendor security advisory: "Does this advisory affect any version we ship?"
• On a pull request that bumps a dependency: "Is the new version free of the findings we have open on the old one?"
• On a package's release notes: "Would upgrading to this release clear our open finding, and does the changelog mention breaking changes?"

The answer combines the page with your own data, so you skip copying package names into another tab.

What leaves your browser

The Chrome Web Store listing is specific about data. Signing in stores an access token, plus your name, email and organization from that token, in local browser storage. Your questions go to Safeguard's API, and so does page content when you have asked the agent to work on that page. The listing states there is no analytics, tracking or telemetry, and that signing out deletes the stored session and revokes the token. Read the listing yourself before rolling the extension out to a team.

Hand Search a tab, with approvals

Some work lives in web tools that have no API: an internal portal, a ticket form, a vendor dashboard. With the Chrome extension, Search can work in a tab you hand it, step by step. You choose how often it stops for your yes.

Three limits apply before you start, all from the store listing and the Safeguard demo:

• Browser control must be enabled for your workspace, and the extension must be signed in. If you cannot find the option, ask your Safeguard administrator.
• It is off by default and turned on per tab. The extension only works in tabs inside its own Safeguard tab group. Your other tabs are off limits.
• You can see when it is working. While a tab is being driven, Chrome shows its notice that the browser is being debugged, and the extension adds a banner on the page that explains what is happening and gives you a way to stop.

Choose a permission mode

Permissions for the browser are Ask, Auto or Yolo, and you set them for each chat.

• Ask stops before every click, every bit of typing and every submit. Use it the first few times, and on any site that holds customer or money data.
• Auto runs reading and navigation on its own and asks before steps such as submitting, paying, deleting or sending. A new chat in Search starts on Auto.
• Yolo runs every step without asking. Keep it for throwaway test sites where nothing real can break.

Separately, the store listing says irreversible actions such as sending, publishing, purchasing, deleting and accepting terms stop and ask first, and that this rule is enforced in the extension's code rather than left to the model. Treat that as a backstop, not a reason to pick Yolo.

Walkthrough: file a ticket in a web portal

This follows the flow in the interactive demo below, which uses a fictional portal and test data.

1. In Search, open the Computers and browsers button under the message box. Check that your Chrome appears under Browser.
2. Set Permissions to Ask or Auto for this chat.
3. Describe the task in plain words, for example: "Open the IT portal and file a ticket asking the platform team to rotate the ci-deploy credentials."
4. Watch it work in the tab you handed it: it opens the portal, reads the page, starts the ticket, types the subject and picks the team.
5. When it reaches the submit step, it stops and asks. Choose Allow once to go ahead or Don't allow to stop. You can answer in Search or in the extension panel.
6. When it finishes, open the step record and check each step and the site it happened on.

Click through the browser control demo to see each of those steps:

Click through it at your own pace, or open it full screen.

How the browser fits with your computer and the cloud

The same Computers and browsers list also shows computers running the Safeguard CLI and the Safeguard cloud. That matters because many tasks cross surfaces: read an advisory in the browser, fix the dependency in a folder on your laptop, check the result on your phone. This recorded desktop session shows the list at "2 connected" with the computer, the browser and Safeguard cloud, then sends work to a folder on the computer:

Click through it at your own pace, or open it full screen.

In that session the CLI session shows in the desktop app's conversation list, you ask for work on your computer, it asks which computer and folder, you pick the folder, then you Approve or Deny each command and file edit. It ends with the upgrade done and tests passing. Our desktop app guide covers that app in depth.

Do the work properly with the Safeguard CLI

The extension is the quickest way to ask. When the answer is "upgrade this package and run the tests", the work belongs on your machine, and that is what the Safeguard CLI is for. It scans code, container images and SBOMs on your own machine, sends only the results to the platform, and includes an assistant that can fix what it finds. It writes code, not only scans.

Install

The CLI is a single static binary with no Node, Python or JVM to install. It runs on macOS, Linux and Windows, on x86_64 or arm64 (Windows is x86_64 only).

# macOS and Linux
curl -fsSL https://cli.safeguard.sh/install | bash

# Windows (PowerShell)
irm https://cli.safeguard.sh/install.ps1 | iex

# Confirm it installed
safeguard --version

The installer detects your OS and verifies the checksum. Our macOS install guide walks through it step by step on a Mac.

First session

1. cd into a project folder and run safeguard with no arguments. The interactive terminal opens.
2. Type /login. Your browser opens; approve the session and the terminal picks it up.
3. Type /scan . to scan the current folder.
4. Ask in plain language, for example "fix the lodash CVE in this repo" or "how many critical vulns do we have".

The assistant has tools for your local files and for your Safeguard data and picks whichever the question needs. So it also works as a general terminal assistant: find and read files ("what's the latest file in ~/Downloads"), edit code and run commands, with your approval.

Writing and fixing code

In the recorded CLI session, the assistant reads the finding, the files and the advisory, then asks before each command. It edits package.json to lodash 4.17.21, runs npm install and npm test with approval, and prints a summary with a suggested commit message. It pushes nothing. You review the diff and commit when you are happy.

Useful extras in the terminal: @ attaches a file from the folder you started in, /model shows or switches the AI model, and /resume picks up the last session in that folder.

Using the plan you already pay for

The plan point at the top of this guide applies here: your existing Claude, ChatGPT or Codex plan can run inside the CLI. For connecting a model, see /model and the CLI docs.

Approvals

Tools that write files, edit code or run commands ask before they act. /permissions in the interactive CLI offers five modes:

There is also a --yolo flag that approves everything. Use it only in throwaway sandboxes.

Scan what changed, and run it in CI

After any agent edits code, yours or another tool's, scan before you merge:

safeguard scan --dir .        # SBOM plus vulnerability match; JSON under ./output
safeguard review              # review working-tree changes for correctness and security
safeguard validate sbom.json  # check a CycloneDX, SPDX or VEX document

For pipelines, safeguard exec "summarise the critical findings in this repository" runs one non-interactive assistant turn. It defaults to strict permissions, so anything the policy file has not pre-approved is denied. Our guide to safeguard exec in CI has full pipeline examples.

When the local runner runs platform scan jobs, your source code never leaves the machine; only SBOMs and findings are uploaded. A CLI session is also live: it shows in the web app, the desktop app and the phone app, and other devices can send work to it. Start in the terminal, carry on at your desk, and approve the next step from your phone.

Troubleshooting

If answers look empty, check you are in the right organization in the web app; answers cover only the scope your account can see.

Common mistakes

1. Using Yolo on real sites. Yolo is for throwaway test pages. On anything that holds customer data, money or access, use Ask or Auto.
2. Handing over a tab with sensitive pages open next to it. Control is limited to the Safeguard tab group, but keep the task's tabs to what the task needs.
3. Expecting browser control in Firefox or Safari. The Firefox listing describes reading a page you attach and says the sidebar clicks nothing. Plan browser tasks for Chrome family browsers.
4. Typing passwords or card numbers into the chat. Sign in to sites yourself, in the tab, before you hand it over.
5. Using the browser for code changes. The panel tells you what to fix; the CLI makes the change on your machine and runs your tests, with your approval.
6. Skipping the step record. After any task that changed something, read the record of steps and the sites they happened on.
7. Assuming Safari on Mac is live. It is in Apple review. Use Chrome, Firefox or the desktop app on a Mac for now.

FAQ

Which browsers does the Safeguard extension support?

Chrome, Brave, Arc, Vivaldi, Opera and Edge through the Chrome Web Store, Firefox through Firefox Add-ons, and Safari on iPhone and iPad inside the Safeguard iOS app. Safari on Mac is in Apple review.

Do I need a separate login for the extension?

You need a Safeguard account. The Firefox listing describes approving the browser from your account, like the CLI sign in, with no password typed into the extension. Signing out deletes the stored session and revokes the token.

Can I keep using my Claude, ChatGPT or Codex subscription?

Yes. Bring the Claude, ChatGPT or Codex plan you already pay for and use it inside the Safeguard CLI. Safeguard adds no extra charge for that usage; your plan's own usage limits and terms still apply. You can also connect Claude to the Safeguard MCP server or ChatGPT, and use Safeguard from the chat you already have open.

Does the extension read every page I visit?

The listings say page content is read and sent only when you ask the agent to work on that page, and the Firefox listing says the sidebar reads a page only when you request it. Questions you type go to Safeguard's API.

Why does Chrome warn about debugging when I install it?

The extension uses Chrome's debugger permission to send genuine input so file pickers and forms work during browser control. Chrome does not let that permission be optional, so the warning appears even if you never turn control on.

What is the difference between Ask, Auto and Yolo in the browser?

Ask stops before every click, typing and submit. Auto runs reading and navigation and asks before steps such as submit, pay, delete and send. Yolo runs every step without asking and belongs on throwaway test sites only.

Are the CLI permission modes the same as the browser's?

No. The browser uses Ask, Auto or Yolo. The interactive CLI offers Ask, Accept edits, Plan, Auto and Strict, plus a --yolo flag for throwaway sandboxes.

Will my side panel conversations show up elsewhere?

Yes. The panel uses the same conversation as the Safeguard web app, so threads carry over to the web, desktop and phone apps.

Try it yourself

This recorded session shows the Safeguard CLI on a Mac: install with the curl command (checksum verified), safeguard --version, then safeguard opens the interactive terminal. You ask "fix the lodash CVE in this repo", the assistant reads the finding, files and advisory, then asks before each command (press y or n). It edits package.json to lodash 4.17.21, runs npm install and npm test with approval, prints a summary with a suggested commit message and pushes nothing. It ends with /permissions, /model and /scan ..

Click through it at your own pace, or open it full screen.

Ready to set it up? Get the extension, the CLI, the desktop app and the phone apps from safeguard.sh/download, and keep the CLI docs open for the first session.

Quick check

Sources

• Safeguard download page
• Safeguard CLI documentation
• Safeguard CLI interactive terminal
• Safeguard CLI slash commands
• Safeguard browser extension docs
• Safeguard local runner docs
• Safeguard on the Chrome Web Store
• Safeguard on Firefox Add-ons
• Chrome Web Store help: install and manage extensions
• Apple support: Safari extensions

Last reviewed: 21 October 2026.