01
Off
Default. The AI sees nothing — SafeCoBrowser is a normal browser.
Built for Claude Code & Codex developers
Local-first · macOSSafeCoBrowser
AI co-browsing, under your control
A macOS browser that lets your own AI agent act on your logged-in sessions — under per-tab, revocable, audited permission. The agent is off by default and never sees more than you grant. Bring your own Claude Code or Codex over MCP or the built-in CLI.
AIAgent & CLI reference — drive SafeCoBrowser over MCP or CLILinux (x64): AppImage · tar.gz
Version 1.0·macOS 14 or later · Linux x64 (Electron)·Bring your own agent
Off by default — the AI sees nothing until you invite it, and Stop AI revokes instantly
Native, not Electron
The Mac App Store build is a rewrite in Swift and SwiftUI — no Electron, no bundled browser engine — for macOS 14 or later. Same product and the same guarantees: the per-tab permission ladder, granting that is never retroactive, instant revoke, approval cards for anything that acts, a hash-chained audit log, and MCP over 127.0.0.1 behind a bearer token.
On Linux, the Electron build above is the one to use — it is maintained and stays the supported option there.
Permission, not access
Everything the agent can do is granted by you, gated per tab, shown before it happens, and written to the log.
Per-tab AI modes: Off, Read, Inspect, Assist, Developer
AI is off by default — no retroactive leak of the blind period
Approve or reject every click, fill, and script
Instant Stop AI — revokes in-flight and future calls
Bring your own agent over a local MCP server or the built-in CLI
Hash-chained, tamper-evident audit log of every action
Isolated containers — a clean room per client or project
Record reusable recipes — replay them, or let your agent read them as how-tos
Privacy filter — redact your name, address, or account numbers from the screen and the agent
Downloads land in each container’s own folder — the agent never sees the files
The agent never gets your cookies, passwords, or profile
Permission modes
Five modes. You decide how far.
Set per tab, by you. Each tier unlocks a little more — and the agent can never raise its own level.
02
Read
Read the page: URL, text, links, and screenshots.
03
Inspect
Inspect elements and read the console and network.
04
Assist
Click and fill — each action behind your approval.
05
Developer
Run JavaScript in the page — full control, always approved.
Trust is the product
Letting an AI act on your logged-in accounts only works if you stay in control. SafeCoBrowser treats that as the default, not a setting.
Off by default
The agent sees nothing until you grant a mode for a tab — and enabling it never exposes what happened while it was blind, like your login.
Instant, real revoke
Stop AI bumps a session epoch that is re-checked at execution time, so in-flight and future calls fail closed the moment you pull access.
Never your profile
The agent only ever gets brokered tools — never the cookie store, saved passwords, or the browser profile on disk.