ego (lite) 只是一個瀏覽器，ego 則是你跨裝置的個人 Agent。
加入候補名單
下載 Mac 版
解決方案
文件企業版更新日誌Blog
ego lite 對比 Codex browser

最好用的 Codex browser 替代品

Codex browser is @Browser in the ChatGPT desktop app: a local Chromium panel Codex can drive, with its own state separate from your Chrome. The Codex CLI has no browser at all.

Run /ego-browser and Codex controls ego lite, the browser you already use, in the CLI and on desktop. Import your Chrome profile and finish whole tasks 3 to 4x faster than Codex browser.

免費試用 ego (lite)GitHub
獲得來自以下單位開發者的信賴
GoogleAmazonShopifyTikTokHarvardStanfordUSCUCLA

為什麼 ego lite 比 Codex browser 更好

You keep Codex. ego lite is which browser it reaches for: your imported Chrome profile with sessions already in place, not a second Chromium you sign into site by site. It works from the CLI where @Browser cannot run, and it beats @Browser on speed, parallelism, tokens, and logins where it can.

在實際任務中快 3 到 4 倍

In the Codex CLI there is no browser at all: the skill may appear, then report unavailable, and you verify changes yourself. On desktop, @Browser re-reads the page through domSnapshot() at every step, and the panel sits idle while Codex thinks.

Run /ego-browser and Codex controls ego lite directly. Every page arrives as a Snapshot: a compressed accessibility tree with stable @N refs Codex can act on instantly. Codex spends its time on the task, not waiting on snapshots or handoffs, finishing whole tasks 3 to 4x faster than Codex browser.

Time to finish a task, shorter is better
81.8 sego lite
282.9 sagent browser
Data source - Task: scrape the main posts from the X account from the last 7 days

平行多工，執行更快

The CLI cannot run browser tasks in parallel because it has no browser. On desktop, @Browser is one panel on one task, and browser work draws from the same usage allowance as your coding work.

In ego lite there's no cap on how many Spaces you open inside the same browser: each one runs its own task at the same time, controlled by Codex or another coding agent via /ego-browser on its own imported Chrome profile. None of them ever steals the tab you're using, and you can watch or take over any Space anytime.

不再來回往返 消耗更少 tokens

On desktop, every @Browser action is its own round trip through the model: domSnapshot(), click, domSnapshot() again. Each result comes back before the next move is chosen, and that spend comes out of the same allowance as your coding work. In the CLI there is no @Browser, so you never get this loop built in.

Run /ego-browser and Codex batches several actions in a few lines of JavaScript per turn to control ego lite. Stacked with Snapshot input, the same task completes on far fewer tokens overall.

Claude Code · ego lite
>/ego-browser Add the first product to the cart, then check checkout loads
●Bash(
ego-browser nodejs <<'EOF' await openOrReuseTab('localhost:3000') await click('.product-card') await click('.add-to-cart') await click('a[href="/checkout"]') cliLog('checkout loaded') EOF
)
⎿Output:
[t=4480ms] checkout loaded
Done. Both steps work.
─── 1 tool call · ↑ 1.1k tokens ───
Codex · @Browser
>Add the first product to the cart, then check checkout loads
●browser.tabs.new() (@Browser)
⎿Tab
Tab created (about:blank)
↑ 1.6k tokens sent so far
Approve localhost for this site, then navigate.
●tab.goto (@Browser)("http://localhost:3000")
⎿Result
Navigated to http://localhost:3000
↑ 2.8k tokens sent so far
Page is up. Reading the DOM before choosing a target.
●tab.playwright.domSnapshot() (@Browser)
⎿DOM snapshot
… +187 lines (ctrl+o to expand)
↑ 4.6k tokens sent so far
First product card is .product-card. Clicking it.
●tab.playwright.locator(".product-card").click() (@Browser)
⎿Result
Clicked .product-card
↑ 6.3k tokens sent so far
Product page rendered. Snapshot again: prior refs are stale.
●tab.playwright.domSnapshot() (@Browser)
⎿DOM snapshot
… +203 lines (ctrl+o to expand)
↑ 8.4k tokens sent so far
Add to cart is .add-to-cart.
●tab.playwright.locator(".add-to-cart").click() (@Browser)
⎿Result
Clicked .add-to-cart
↑ 10.2k tokens sent so far
Added. Opening checkout, then confirming the form with a snapshot.
●tab.playwright.locator("a[href='/checkout']").click() (@Browser)
⎿Result
Clicked a[href='/checkout']
tab.playwright.domSnapshot() → payment form present
↑ 13.6k tokens sent so far
─── 7 tool calls · ↑ 13.6k tokens ───
同樣的任務，同樣的模型。 左邊：ego lite 裡一次打包的 JavaScript 呼叫，總共 1.1k tokens。 右邊：Codex 內置瀏覽器，7 次往返，token 漲到 13.6k。

同一個 Chrome，AI Agent 原生體驗

In the CLI, Codex never sees your Chrome logins because there is no browser. On desktop, @Browser uses separate state: OpenAI's help center says it does not use your Chrome profile, cookies, tabs, or extensions. You sign into a second browser by hand, one site at a time.

Built on Chromium, ego lite imports your entire Chrome setup in one click. When Codex runs /ego-browser, it controls ego lite with your real sessions already in place, including the 2FA trust you already have, so it doesn't stall on login walls you passed months ago.

ego lite Chrome 個人設定檔匯入：一鍵設定，帶入你所有的登入狀態

ego lite 對比 Codex browser

ego lite 與 Codex browser 的功能比較。
功能ego liteCodex browser
可在 Codex CLI 中使用Yes. Run /ego-browser and the coding agent controls ego liteNo. Only the desktop app has @Browser
沿用你真實的 Chrome 登入狀態Yes, one-click Chrome profile importNo. Separate browser state, signed in by hand per site
跨源 iframe 與 passkeySnapshot reaches into cross-origin iframes and shadow DOMOpen issues: cross-origin iframe input unsupported; passkeys incomplete
AI Agent 如何在頁面上執行動作Coding agent batches multiple actions in JavaScript per turn via /ego-browserOne tool call per action, result read back each time
每項任務的 Token 成本Lower: Snapshot plus batched actions, measured per task更高：每一步都要把上下文帶回模型
工作按什麼計費不需要額外東西。瀏覽器免費且在本地The same coding allowance as the rest of your Codex work (5-hour window plus a weekly cap)
平行任務隔離Unlimited Spaces in one browser, bounded by your machine每個會話一個內置瀏覽器面板
同時可作為你日常使用的瀏覽器可以，你在自己的 Space 裡瀏覽，Agent 在它們自己的 Space 裡工作No. A separate panel with separate state
與真實帳號隔離No. The coding agent controls ego lite inside your loginsYes, by design: separate state, no path to personal accounts
可重複使用的 skills（即將推出）將成功的執行過程萃取成可重複使用的 skills；隨著 AI Agent 重複執行，複雜任務最多可加快 5 倍（限量 beta）沒有內置的瀏覽器任務等價物
價格與可用性Free, no subscription. macOS, Chromium onlyIncluded with ChatGPT plans; Computer Use is region-limited
最後更新 2026年7月30日

讓轉換無縫接軌

You are not switching away from Codex. Run /ego-browser so Codex controls ego lite, a Chromium browser that already holds your sessions, from the CLI or alongside @Browser on desktop.

  1. 下載 ego (lite)

    下載 ego lite，一鍵匯入你的 Chrome 設定檔，登入狀態、Cookie 和擴充功能全部一併帶入。

  2. 用 /ego-browser 執行你的第一個任務

    貼到你的 Agent 裡

    /ego-browser 前往 ego.app 並加入候補名單

    Run /ego-browser in Codex and paste one sentence. Works in the CLI; no MCP config, no per-site approval step first.

  3. 看它實際運作
    ego lite 的 Space 總覽，四項瀏覽器任務並排執行：Claude Code 在 Yahoo Finance 追蹤蘋果股價、Codex 在 cars.com 依年份篩選車款、Hermes 完成一項 SaaS 後台任務、一位使用者在擷取 X 上的資料，還有一隻手正點擊 + 開啟另一個 Space

    The task runs in its own Space, so your tabs stay untouched. Take over anytime. Results land back in the Codex CLI.

Keep @Browser on desktop for localhost previews and the annotation loop: select an element, comment, fix exactly that.

何時該用哪個工具

適合選擇 ego (lite) 的情況

  • You want browser work from the Codex CLI, where @Browser does not run.
  • Logged-in tasks should start in your real Chrome, not a second browser you authenticate site by site.
  • Even on desktop, whole tasks behind SSO feel slow in @Browser's one-panel, one-action loop.
  • You want parallel Spaces without spending Codex allowance on routine @Browser clicking.

選擇 Codex browser 的情境

  • You want zero setup: @Browser is already in a ChatGPT plan you pay for.
  • You're debugging your own frontend on localhost and want CDP access from inside the agent.
  • You want the annotation loop: select an element, comment, fix exactly that.
  • You want the agent to have no path to your real accounts. Separate state is a deliberate safety tradeoff.

給你的 AI Agent 一個真實瀏覽器

免費，在你的 Mac 上執行，一鍵匯入你的 Chrome 個人設定檔。適用於 Claude Code、Codex、Cursor，以及任何會寫程式碼的 CLI AI Agent。

立即試用 ego (lite)

繼續閱讀

還在權衡？看看 Codex browser 和同賽道其他工具怎麼比。

ego (lite) vs Puppeteerego (lite) vs Playwright MCPego (lite) vs Browser Useego (lite) vs Browserbaseego (lite) vs Stagehandego (lite) vs agent-browserego (lite) vs Playwrightego (lite) vs Selenium所有比較瀏覽器自動化解決方案

FAQ

Codex browser is OpenAI's built-in browser for Codex, reached with @Browser in the ChatGPT desktop app. It is a local Chromium panel you can watch and take over, with DOM and accessibility-tree reading plus CDP access in Developer mode. For localhost frontend work it is genuinely good, and it is included in ChatGPT plans. It does not use your Chrome profile, and it is not available in the Codex CLI. The Codex Chrome extension is a separate surface covered on its own compare page.

Yes, for both surfaces. In the Codex CLI, ego lite is the Chromium browser Codex lacks. On desktop, it beats @Browser on speed, parallelism, tokens, and logins because Codex controls ego lite with batched JavaScript instead of one domSnapshot round trip per click. Keep @Browser for localhost previews and the annotation loop.

Codex browser use is OpenAI's @Browser feature in the ChatGPT desktop app: Codex can open pages, click, and inspect them in a local Chromium panel. The Codex CLI has no browser. Run /ego-browser and Codex controls ego lite, a Chromium browser with your imported Chrome profile, from the CLI and beats @Browser on whole logged-in tasks.

OpenAI Codex browser is the built-in Chromium panel reached with @Browser in the ChatGPT desktop app. It is local, not cloud, and separate from your Chrome profile. ego lite is not a replacement for Codex itself. It is the Chromium browser Codex controls when you run /ego-browser from the CLI or want your real Chrome sessions on desktop.

In the ChatGPT desktop app, yes, via @Browser for codex web browser tasks. In the Codex CLI, no: only web search snippets, no page control. Codex cloud blocks internet by default in the agent phase. Run /ego-browser so Codex controls ego lite, a Chromium browser, from the CLI.

No. The browser skill may still appear, then report unavailable. Run /ego-browser so Codex controls ego lite, a Chromium browser on your Mac, from the same terminal. On desktop, @Browser exists but uses separate state and a slower one-action loop; ego lite still wins for logged-in whole tasks.

On desktop, invoke @Browser and Codex drives its built-in panel one action at a time: tab.goto, domSnapshot, locator click. From the CLI, install ego lite, import your Chrome profile, and run /ego-browser so Codex controls ego lite against your real sessions with batched JavaScript instead of seven round trips per task.

Yes, into its own browser state, not your Chrome profile. OpenAI's help center points to the Codex Chrome extension when you need existing cookies or sessions. ego lite imports your profile once, so Codex starts inside the logins and 2FA trust you already have.

No. @Browser is an embedded Chromium panel in the desktop app on your machine. ChatGPT's separate cloud browser is a different surface. ego lite is also a local Chromium browser on your Mac: free, no subscription.

Yes. @Browser draws on the same coding allowance as the rest of your Codex work. ego lite adds no meter of its own; you only pay model tokens your agent already spends, and batched JavaScript keeps per-task totals lower than @Browser's one-action loop.

Playwright MCP works in the CLI but needs MCP config and one tool call per action against a blank profile. ego lite needs no MCP: run /ego-browser against your real Chrome profile from Codex, Claude Code, or Cursor. See ego lite vs Playwright MCP.