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 vs Browser Harness

The Best Browser Harness Alternative

Browser Harness is Browser Use's local version: a Python agent loop you run on your own machine with your own LLM API key, launching its own browser and calling the model again at every step to plan the next move.

ego lite needs no framework: the coding agent you already run drives your real, logged-in browser directly, batching actions instead of narrating them one at a time. Across 31 live-site jobs it finishes 96.8% of tasks with a perfect score to Browser Harness's 83.9%, on 41% fewer model round trips.

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

Browser automation benchmark: ego lite vs Browser Harness

This is live browser automation, not a sandbox demo. ego lite and Browser Harness, Browser Use's local version, ran the same 31 multi-step tasks on real websites: several pages, several decisions, often a login you already have. The hosted cloud product was not part of this run. Local against local, same jobs, scored the same way.

agent: pimodel: ChatGPT 5.6 Solthinking: max31 tasks

Task completion rate

How often the agent finished the job. Stuck at login, skipped a step, or returned the wrong result: that task is a fail. Higher is better.

ego lite96.8%30 of 31 tasks perfect
Browser Harness (Browser Use local version)83.9%26 of 31 tasks perfect

Cost per completed task

What one finished job costs in model spend. Failures still get billed, so they push this number up. Lower is better.

ego lite$1.98
Browser Harness (Browser Use local version)$3.04

Model turns per task

How many times the model had to look at the page and choose the next action. Extra looks mean extra tokens, extra waiting, and extra places to stall. Lower is better.

ego lite30.3
Browser Harness (Browser Use local version)51.2

Average task time

How long a job took from start to finish, on average, including time spent waiting on the model. Lower is better.

ego lite8m 38s
Browser Harness (Browser Use local version)9m 58s

We ran every tool twice and kept the better score. A failed task still counts against it.

Check the numbers, or rerun the tasks yourselfThe 31 tasks, the grading checklists, and the raw results are open source. If a number looks off, open the repo.ego-browser-benchmark-framework

Why ego lite is better than Browser Harness

With Browser Harness, the browser agent is something you build: a Python project, an LLM API key, a loop that calls the model every step. With ego lite, it's something you already have. Run /ego-browser inside Claude Code, Codex, or Cursor and your agent gets your actual browser, logins and all. (Browser Use's hosted cloud product is a different category; the local-versus-local matchup is this page.)

96.8% of tasks done perfectly. No other tool tops 84%.

Browser Harness's agent loop calls the LLM at every step: read the page, decide the next batch of actions, wait, repeat. It turned in the most careful runs of any competitor on those 31 jobs, and paid for the care in round trips: 51.2 model calls per task, the most of the five tools measured, each one re-sending page state just to plan the next few clicks.

In ego lite, every page reaches your agent as a compressed Snapshot it can act on immediately, several actions per JavaScript turn, no separate model loop deciding a step at a time. The result on the same 31 tasks: 96.8% perfect to Browser Harness's 83.9%, on 41% fewer round trips and 35% less model spend per completed task.

Time to finish a task, shorter is better
518 sego (lite)
598 sbrowser harness
Data source - Task: finish an average live-site job

平行多工，執行更快

Browser Use's own docs mark parallel Browser Harness runs experimental: developers note agents can conflict when they share resources, and real isolation means standing up a separate browser instance per task yourself.

在 ego lite 中，Space 數量沒有上限：每個 Space 都能同時執行自己的任務，使用各自匯入的 Chrome 瀏覽器設定檔。它們都不會搶走你的分頁，你也能隨時觀看或接手任何一個 Space。

不再來回往返 消耗更少 tokens

Browser Harness's loop calls the model at every step and carries the page state along, and on longer tasks that cost compounds fast enough that users ask the Browser Use maintainers how to avoid burning tokens on a task that runs the same way every time.

在 ego lite 中，你的 Agent 每一輪只需幾行 JavaScript 就能批次執行多個動作，不必等待新的模型呼叫才能繼續。再加上 Snapshot 輸入的加乘效果，同樣的任務整體所耗費的 Token 大幅減少。

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 ───
Browser Harness · agent.run()
>$ python run_task.py 🎯 Task: Add the first product to the cart, then check checkout loads
📍 Step 1:
👍 Eval: Page loaded successfully
🧠 Memory: On demo store homepage, need to find a product
🎯 Next goal: Open first product and add it to cart
📍 Step 1: Ran 3 actions in 2.1s: ✅ 3
↑ 3.4k tokens sent so far
📍 Step 2:
👍 Eval: Product added to cart
🧠 Memory: Item in cart, now check checkout
🎯 Next goal: Navigate to checkout and confirm it loads
📍 Step 2: Ran 2 actions in 1.9s: ✅ 2
↑ 7.1k tokens sent so far
📍 Step 3:
👍 Eval: Checkout page did not load on first click
🧠 Memory: Link required a second click, retrying
🎯 Next goal: Retry the checkout link and wait for the form
📍 Step 3: Ran 1 action in 2.3s: ✅ 1
↑ 10.8k tokens sent so far
📍 Step 4:
👍 Eval: Payment form is present
🧠 Memory: Task complete
🎯 Next goal: done
📍 Step 4: Ran 1 action in 1.4s: ✅ 1
↑ 13.6k tokens sent so far
─── 4 model calls · ↑ 13.6k tokens ───
The same task, the same model. Left: one batched JavaScript call in ego lite, 1.1k tokens total. Right: Browser Harness's agent loop, a fresh model call per step, tokens climb to 13.6k.

同一個 Chrome，AI Agent 原生體驗

Browser Harness launches its own Chromium over the DevTools protocol, and the Browser Use GitHub issues are full of developers hitting profile-lock errors pointing it at their real Chrome: the --profile flag copies the live profile into a temp directory, and on Windows that copy fails outright while Chrome holds the file locks.

ego lite 建構於 Chromium 之上，能一鍵即時匯入你完整的 Chrome 設定，不需要先關閉任何東西。你的 Agent 能直接繼承真實的登入狀態，不會卡住。

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 vs Browser Harness

Feature comparison between ego lite and Browser Harness.
功能ego liteBrowser Harness
由誰提供模型你的程式設計 Agent（Claude Code、Codex、Cursor）你自己：自備 LLM API 金鑰，並依 Token 用量付費
設定安裝應用程式，在你的 AI Agent 中執行 /ego-browserPython 專案、pip 安裝、模型設定、任務程式碼
已登入的網站（SSO、2FA）一鍵匯入 Chrome 瀏覽器設定檔，預設已登入啟動的是全新 Chromium，設定檔需手動重複使用，且有諸多限制
動作如何執行每輪用 JavaScript 批次執行多個動作Agent 迴圈每一步都呼叫模型
每項任務的 Token 成本較低：Snapshot 輸入加上批次動作，以整項任務計算較高：每一步都要透過模型重新傳送頁面狀態
Browser automation on live sites (31 tasks, same model)96.8% perfect, $1.98 per completed task, 30.3 round trips83.9% perfect, $3.04 per completed task, 51.2 round trips
平行任務Space 在同一個可見的瀏覽器中隔離各項任務實驗性功能：需要多個 Agent 實例，或使用付費雲端方案
同時可作為你日常使用的瀏覽器可以，你在自己的 Space 瀏覽，Agent 則在它們自己的 Space 中工作不行，這是自動化函式庫加上託管瀏覽器
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）沒有內建的對應功能
價格免費，無需訂閱開源版本免費；雲端與 LLM 用量須付費
最後更新 2026年8月20日

讓轉換無縫接軌

If you set up Browser Harness to automate your own browsing (research, form filling, logged-in chores) rather than to ship a product, the switch removes the whole project layer.

  1. 下載 ego (lite)

    Download ego lite and import your Chrome profile in one click. The logins you were configuring Browser Harness to reach come along automatically.

  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

    在 Claude Code、Codex 或 Cursor 中執行 /ego-browser。不需要 Python 環境，不需要選擇模型，也不需要 API 金鑰。

  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

    任務會在自己的 Space 中執行，而不是只能從記錄檔猜測狀況的無頭 Chromium 實例。你可以即時觀看，或隨時接手，結果最終會回到你 Agent 的 CLI 中。

Keep Browser Harness where it belongs: inside Python products and pipelines you're building for others. ego lite covers the agent browsing you do yourself.

何時該用哪個工具

適合選擇 ego (lite) 的情況

  • 你想要有個 Agent 幫你處理自己的瀏覽工作（研究、填表單、已登入的雜務），而不需要用 Python 建構任何東西。
  • 你已經在使用 Claude Code、Codex 或 Cursor，不想再多一個需要自付 API 金鑰費用的 Agent 迴圈。
  • 任務需要你真實的登入狀態：一鍵匯入設定檔勝過手動設定 Chromium 設定檔。
  • 你希望任務能在多個平行的 Space 中執行，並在一個可即時觀看或接手的有頭瀏覽器中進行。

Choose Browser Harness when

  • You're building a custom automation product or pipeline in Python. Browser Harness is a library designed to be embedded.
  • 你想在自己的程式碼中完全掌控 Agent 迴圈：自行選擇模型、設計提示詞、控制每一個步驟。
  • You need automations deployed to the cloud, running on Browser Use's hosted browsers when your laptop is closed.
  • 你要把瀏覽器自動化交付給其他使用者。ego lite 是一個終端使用者瀏覽器，而不是 SDK。

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。

免費用 ego (lite)

繼續閱讀

Still weighing your options? See how Browser Harness compares with the other tools in the same space.

ego (lite) vs Puppeteerego (lite) vs Playwright MCPego (lite) vs Claude Chrome extensionego (lite) vs Browserbaseego (lite) vs Stagehandego (lite) vs agent-browserego (lite) vs Playwrightego (lite) vs Seleniumego (lite) vs Codex browserego (lite) vs Kimi Chrome extensionego (lite) vs Codex Chrome extension所有比較瀏覽器自動化解決方案

FAQ

Browser Harness is Browser Use's local version. Browser Use is one of the most popular open-source AI browser agent projects: an MIT-licensed Python agent loop that lets an LLM control a browser, launching its own Chromium over the DevTools protocol and deciding actions step by step against the page state. You bring your own model API key and run it on your own machine. Browser Use's paid cloud product runs hosted browser agents with stealth and captcha-solving features, which is a different category from a local browser, so this page compares the harness: local against local. For developers building custom automation products and pipelines in Python, it's a strong, actively developed foundation, and of the five local tools we measured it posted the best completion rate of any competitor.

If you want an agent to do browser work for you, yes. ego lite plugs your existing coding agent into your real logged-in browser with no Python project and no API-key billing, and across 31 live-site jobs it finished 96.8% of tasks perfectly to Browser Harness's 83.9%, at $1.98 per completed task against $3.04. If you're building a browser-automation product or pipeline in Python, Browser Use is the better fit: it's a library designed to be embedded, and ego lite isn't an SDK.

Browser Use is the project; Browser Harness is its local version, the open-source Python agent loop you run on your own machine with your own LLM API key. Browser Use also sells a cloud product that runs hosted browser agents on their infrastructure. ego lite is a local browser, so the apples-to-apples comparison on this page is against Browser Harness; if you're weighing hosted browser infrastructure, see our Browserbase page instead.

We ran both tools on the same 31 browser automation tasks on live websites, driven by the same pi agent with the same model (gpt-5.6-sol, thinking effort max) and graded by the same written checklist. Each tool keeps its better of two full runs, so a failed task still counts. ego lite finished 30 of 31 tasks perfectly (96.8%) at $1.98 per completed task on 30.3 model turns; Browser Harness finished 26 of 31 (83.9%) at $3.04 on 51.2 turns, the best completion rate of any competitor measured. Browser Use's hosted cloud product was not part of the run; the comparison is local against local. The tasks, the checklists, and the raw results are open source in the ego-browser-benchmark-framework repository on GitHub, so every number can be audited or reproduced.

Browser Harness can, with manual configuration, and it's a common source of friction. The --profile flag copies your live Chrome profile into a temp directory, and on Windows that copy fails outright while Chrome is running and holding file locks, so the workaround is closing Chrome first. Connecting to an already-running Chrome over CDP works too, but Chrome 136 and later blocks CDP on your default profile, so most setups end up on a separate, non-default one. In ego lite, the real profile is the starting point: one-click import of logins, cookies, sessions, and extensions, shared safely with your own browsing through separate Spaces.

不需要。這裡沒有另一個需要你付費的模型迴圈。智慧的部分，來自你原本就在使用的程式設計 Agent，不論是 Claude Code、Codex、Cursor、Gemini CLI 還是 OpenCode。ego lite 本身完全免費，也不需要訂閱。

兩者都是以 LLM 驅動瀏覽的開源工具，但 Browser Use 是圍繞自主迴圈打造的 Python Agent 框架，而 Stagehand 則是 Browserbase 推出、以 TypeScript 為主的 SDK，提供 act / extract / observe 等基本操作，用於程式碼加 AI 混合式自動化。兩者都是以程式碼優先：你得先建立專案架構，再提供模型存取權限。想了解另一側的比較，可參考我們的 ego lite 與 Stagehand 比較頁面。

它們解決的是不同的層級。Browser Use 是 Agent 框架；Browserbase 則是供框架執行的雲端瀏覽器基礎設施。團隊經常會把 Agent 函式庫與託管瀏覽器搭配使用，以應付正式環境的規模，這對雲端叢集來說確實是個好組合。ego lite 不屬於這兩個層級中的任何一個：它是一個本機的有頭瀏覽器，讓你在自己的電腦上、用自己的帳號，完成 Agent 的工作。

Browser Harness's agent loop calls the model at every step, and each call carries a fresh serialized snapshot of the page just to plan the next batch of clicks. Users running recurring or long-horizon tasks have asked the maintainers how to cut that cost, since the same task run the same way still pays full model price every time. On those 31 jobs that loop averaged 51.2 model round trips per task, the most of the five tools measured, against ego lite's 30.3. ego lite reduces the round trips themselves: the agent batches several actions in one JavaScript execution and reads pages as compressed Snapshots, so the savings compound over a whole task.

負責把 Agent 連接到瀏覽器的 ego-browser 外殼，是採用 MIT 授權的開源專案。而 ego lite 瀏覽器本身，則是一款適用於 macOS 的免費應用程式，不需訂閱，你的資料也都留在本機。