Guide
ego lite Explained, and How It Compares with Aside
ego lite and Aside are both browsers built for AI agents, but one has no agent of its own and the other ships with one. A guide to what each is, with the same page-reading test run on both.
If you let a coding agent use a browser, the default is usually a blank, headless one. It has no logins, no cookies, and every site greets it with a sign-in page. ego lite and Aside both exist to fix that, and they fix it from opposite ends. ego lite is a browser with no AI inside it, built so that agents you already use (Claude Code, Codex, Cursor) can borrow it. Aside is a browser with its own agent inside, which can also be driven from the outside. I installed both on one Windows 11 machine and ran the same small test on each, so this guide mixes what the vendors say with what I saw.
What ego lite is
ego lite is a free, Chromium-based browser from Citro Labs. Its README describes it as a browser where you and your AI agents work in parallel.1 Three ideas matter.
Task spaces. Each agent job gets its own isolated workspace, called a Space, with its own tabs. Your own tabs are not touched. You can see which Space has an agent running, take it over, or stop it.1
Your logins come along. On first launch it offers to import your Chrome data, and the README says the agent then inherits your logins, cookies, extensions and bookmarks.1 There is also a command for this: ego-browser import --browser chrome --profile Default.
The agent writes a script. There is no model inside ego lite. The agent you already run writes a short JavaScript program against a small API (goto, snapshot, click, fill, waitForSelector) and sends it to ego-browser nodejs. One script can do many steps and print one final snapshot, instead of one tool call per click.12 The API is deliberately not Playwright. The bundled usage guide tells the agent not to invent Playwright methods like locator() or getByRole().2
A snapshot is a text view of the page, with short references like @21 that later actions can point at. That is how a model “sees” a page without a screenshot.
On my machine: ego lite 0.6.1.13 on Chromium 156, Windows 11. The README lists macOS and Windows as supported and Linux as planned.1
What Aside is
Aside is also a Chromium-based desktop browser, but it ships with an agent. Its site says it gets work done “across your websites, accounts, and history.”3 Its own pieces:
- Memory. It turns browsing history into notes so you do not repeat context.3 The CLI can search them (
aside memory search). - A password manager (Aside Vault). The site says agents sign in through autofill and the credentials are not exposed to the agent, and that sensitive actions wait for human approval.3
- Three ways to drive it from a terminal.
aside exechands a task to Aside’s own agent, like spawning a subagent.aside replruns Playwright-style JavaScript directly.aside mcpexposes it as an MCP server.4
So an outside agent (say, Claude Code) has a choice that does not exist in ego lite: delegate the whole task to Aside’s agent, or use Aside only as a remote-controlled browser. The CLI guide recommends exec for most tasks and repl only when you need to look at the screenshot and DOM yourself.4
Windows support is real but newer than macOS. The changelog shows macOS first (v1.0.225.1, February 2026) and Windows in v1.0.914.1, whose entry reads “Aside is now officially available on Windows!”5 I covered that gap in a post on the cost of standalone browsers. On my machine: Aside app 1.0.1022.1 and CLI 1.26.916.1741.
flowchart LR subgraph E["ego lite"] EA["Your agent<br/>(Claude Code, Codex...)"] -->|"JS script"| EB["ego-browser<br/>task space"] EB -->|"snapshot"| EA end subgraph S["Aside"] SA["Your agent"] -->|"repl / mcp"| SB["Aside browser"] SA -->|"exec: whole task"| SG["Aside's built-in agent"] SG --> SB SG --> SM["Memory +<br/>password manager"] end
Where they differ
| ego lite | Aside | |
|---|---|---|
| Built-in agent | No. Bring your own | Yes, plus memory and a password manager |
| How outside agents drive it | ego-browser nodejs scripts |
exec, repl, or MCP |
| Isolation | Each job in its own Space, by design | REPL can list and attach to your open tabs; external agent tabs are grouped in the sidebar45 |
| Model choice | Whatever your agent uses | Set per task with -m, --effort, and a permission mode (ask/guard/full-access)4 |
| Page snapshot | Viewport or scope: "full_page" |
Accessibility tree, capped at 16 KiB (below) |
| License | Skill and docs are MIT; the app is a separate free download1 | Closed product, bring your own subscription or API key3 |
The isolation row is the one I would think hardest about. ego lite’s default is to keep the agent away from the tabs you are using. In Aside’s REPL, listBrowserTabs() returned all 23 tabs I had open, and the guide says to use attachActiveBrowserTab() when you want the agent to work on the page you are looking at.4 That is convenient and also a larger blast radius. Neither is wrong, but it is a decision to make on purpose.
The same test on both
I wanted one number that was not a vendor benchmark. The test: open a URL, then read it as a snapshot. Three pages, five runs each, one machine, one network. Scripts are in the repo.6
example.com(tiny)lleg.devhome page (this blog)- the English Wikipedia article “Web browser” (large)
Two caveats come before the numbers. The two tools do not wait for the same thing. ego’s goto reuses one page and waits for the load event. Aside’s openTab creates a new tab each time and waits until the page is interactive and the DOM is stable. So “open” time is not an apples-to-apples speed contest, and I did not try to make it one.
Show data table
| page | ego lite | Aside |
|---|---|---|
| example.com | 211 | 3,928 |
| lleg.dev | 333 | 2,443 |
| Wikipedia | 501 | 3,678 |
ego was faster to open on all three, by roughly 7 to 19 times. Variation was large on both sides: Aside’s example.com runs ranged from 2.9 to 7.6 seconds, and one Wikipedia run took 15.2 seconds; ego’s first run on each page was the slowest (up to 2.6 seconds) and the rest were close to the median. I do not know how much of Aside’s gap is tab creation and stability waiting versus the browser itself, so I would not quote the ratio without that caveat.
The snapshot step is more interesting, because it shows a real design difference.
| Page | ego full page | ego viewport | Aside default |
|---|---|---|---|
| example.com | 1,176 chars | 1,176 | 265 |
| lleg.dev | 10,118 | 3,992 | 5,374 |
| Wikipedia | 306,375 | 16,287 | 16,437 |
Time to read the snapshot was small everywhere except one case: ego’s full-page Wikipedia snapshot took a median 8.2 seconds, against 0.8 seconds for Aside’s.
That is not Aside being faster at the same job. Aside’s Wikipedia snapshot ended with [snapshot truncated at 16 KiB; narrow it with { interactive: true }, selector or ref]. It returned the first 411 lines, and ego’s full-page snapshot of the same page was 4,668 lines. Both are reasonable defaults. Aside’s keeps each read small enough to fit a model’s context and asks you to narrow the query. ego’s full_page gives you everything and leaves the budgeting to the agent, which on a long article means a few hundred thousand characters, roughly tens of thousands of tokens if you pasted it into a model whole. The default viewport snapshot (16 KB here) is the cheap path.
What I did not measure: aside exec (Aside’s own agent), end-to-end task success, token cost, anything behind a login, or Linux and macOS behavior. One of my first Aside runs hung for 34 minutes and left the browser disconnected from its CLI until I restarted it, so there is a robustness note too, though I could not reproduce it on the second attempt.
What the vendors claim
Both publish benchmark numbers, and they cannot be compared with each other.
ego lite’s README says its tool completes complex web tasks “47% to 63% faster than the other tools” on 31 live-site tasks, run with the vendor’s choice of agent and model.1 Aside’s site says it ranks first on three agent benchmarks, with 99.0% on Online-Mind2Web against 97.7% for Browser Use.3 These are different benchmarks, run by the companies selling the product, measuring different things (speed of a tool versus success of a built-in agent). I could not verify either, and neither README nor site links raw per-task results I could check.
On privacy, Aside says tasks and memory are local by default and that “nothing is shared with LLM providers.”3 The same page offers bring-your-own subscription, which sends requests to a provider, and does not reconcile the two statements. ego lite says browsing data and cookies stay on the device.1 But note what that covers: when your agent reads a snapshot, the page text goes to whichever model is behind that agent. That is true for any agent browser and is my inference, not a claim from either vendor.
Which one to pick
- You already work in Claude Code or Codex and want it to use sites you are logged into. ego lite is the smaller step. It adds a browser and nothing else, and Spaces keep the agent out of your tabs.
- You want to hand a whole chore to an agent and read the result. Aside’s
exec, memory and password manager are built for that. I did not testexec, so try it on something low-stakes first. - You care most about token budget. Look at the snapshot defaults before anything else. Aside caps a read; ego lets you choose viewport or full page.
- You are on Linux. ego lite lists it as planned, not shipped.1 For Aside I could only confirm macOS and Windows, so check its download page.
- You will run unattended jobs. Test a long loop first. My one hang with Aside is a single observation, not a verdict.
References
Footnotes
-
Citro Labs, ego lite README, read 2026-10-28. Source of Spaces, Chrome data import, platforms, MIT license of the skill repo, and the benchmark claims (31 tasks, 47% to 63%), which are the vendor’s own. ↩ ↩2 ↩3 ↩4 ↩5 ↩6 ↩7 ↩8 ↩9
-
ego-browser skillusage guide v2.1.0, printed by the installed 0.6.1.13 on 2026-10-28. Source of the script-based API and the instruction not to assume Playwright methods. ↩ ↩2 -
Aside, homepage and Help Center, read 2026-10-28. Memory, Aside Vault, “local by default”, bring-your-own subscription and the benchmark figures are vendor statements. ↩ ↩2 ↩3 ↩4 ↩5 ↩6
-
aside guideandaside guide repl,aside --help, CLI version 1.26.916.1741, run 2026-10-28. Source ofexec/repl/mcp, model and permission flags,listBrowserTabs()andattachActiveBrowserTab(), and the 16 KiB snapshot truncation message (observed directly). ↩ ↩2 ↩3 ↩4 ↩5 -
Aside, native app changelog. v1.0.225.1 initial release, v1.0.914.1 Windows availability and the “Agent Tabs” sidebar group for tabs used through MCP or CLI. ↩ ↩2
-
Methodology. 2026-10-28, Windows 11, one machine, one network, five runs per page, medians reported. Scripts:
ego-vs-aside-bench-ego.mjsandego-vs-aside-bench-aside.js. ego usedpage.gotothensnapshot({ scope: "full_page" })andsnapshot(); Aside usedopenTabthensnapshot(page, {}). Aside was run one URL per call because the REPL has a 120-second limit. Wikipedia full-page and truncation line counts come from separate single checks. ↩