BrowserSkill alternative
BrowserSkill, from Tencent, lets the agent in your terminal use your logged-in Chrome or Edge, in a window of its own. Browsentic runs the same kind of agent from a side panel beside your tab, in Firefox as well, with voice, schedules and approvals. Both keep your logins on your computer.
Where a job starts
With BrowserSkill you ask in your agent’s terminal, and it works in a window of its own. With Browsentic you ask in the side panel, and Browsentic Bridge starts your agent to work in the tab beside it.
BrowserSkill
Your agent’s terminal
Compare these three laptops and keep the one under €1,200
Your agent
Reads the browser-skill skill, then runs bsk commands
Claude Code Codex Cursor OpenClaw CodeBuddy
The bsk command line and its daemon
A Rust binary on your computer
A separate Agent Window
One of your own tabs only when you let it borrow one
Browsentic
Side panel, beside your tab
Compare these three laptops and keep the one under €1,200
Browsentic Bridge
On 127.0.0.1, it starts your agent as you
The agent CLI you already have
Claude Code Codex Antigravity Cursor CLI
The tab beside the panel
Each step on the panel’s timeline; consequential ones wait for you
Browsentic can be driven from a terminal as well: Browsentic Bridge also serves MCP, so Claude Code, Cursor or Zed can use the same browser. BrowserSkill has no side panel, so every job starts from an agent.
At a glance
A command line and an extension for the agent in your terminal
An agentic browser harness with a side panel beside your tab
Side by side
Checked against the BrowserSkill repository, its bundled skill and its store listings, 14 October 2026. If something here is out of date, open an issue.
| Browsentic | BrowserSkill | |
|---|---|---|
| The shape of it | ||
| What it is | A side panel in your browser, plus Browsentic Bridge on your computer | The bsk command line and daemon, an extension, and a skill installed into your agent |
| Where you ask | In the side panel, typed or spoken; or from an MCP client | In your agent’s terminal or app |
| The agent | Starts the CLI you already have: Claude Code, Codex, Antigravity, Mistral Vibe, Grok Build, Cursor CLI, Qwen Code or OpenCode | Any agent with a shell that has the skill: Claude Code, Codex, Cursor, OpenClaw, CodeBuddy, WorkBuddy, Pi or Hermes Agent, and a DeepSeek Harness plugin |
| Where it acts | The tab beside the panel, and the tabs it opens | A separate Agent Window; one of your tabs only after you confirm |
| Model keys | None | None |
| License | Apache 2.0 | MIT |
| Latest release | v0.8.9, on 12 October 2026 | Command line v0.3.2 on 30 September 2026, extension v0.3.3 on 14 October 2026 |
| Where it runs | ||
| Browsers | Chrome, Edge, Brave, Arc and other Chromium browsers, and Firefox as a signed add-on | Chrome and Edge, version 125 or newer; other Chromium browsers may work, without support; no Firefox build |
| Installing it | One line or an app installs Browsentic Bridge and opens the store listing; pair once with a code | One line installs bsk, the extension comes from its store, then bsk install-skill for each agent |
| A desktop app | Mac and Windows apps that install, update and run everything | None; the command line updates itself |
| Agent on another machine | No: everything binds to 127.0.0.1 | Yes: an agent on a server pairs with your browser over authenticated WSS |
| A phone | Chrome on an Android phone, experimental | No |
| What it can do | ||
| Voice | Hands-free: talk to it, or hold a key to talk | No |
| On a schedule | Scheduled tasks, each in a background tab, and monitors that watch a page for you | No |
| Show it a job once | Recordings: do it yourself, later say “do it like last time” | No; its recordings are silent videos of a tab |
| “I am not a robot” checks | Solves them on the way | Asks you to step in, as for a login or a one-time code |
| Debugging a website | Reads network requests and console output during a run | Ties each action to its requests, responses and console output; mocks, blocks or replays requests; exports the evidence |
| Tables | Read as text | Exported as JSON or CSV, with column names |
| A site’s own tools | Calls the WebMCP tools a site registers for agents | No |
| Trust | ||
| Before it commits | Approval in the panel before submitting, uploading or leaving the run’s sites, tunable per site | Confirmation before it borrows one of your tabs |
| Your passwords | Sealed into placeholders: the agent types them without reading them | Its skill tells the agent never to extract credentials; known secrets are filtered from debugging evidence |
| Sites kept off limits | A blocked-sites list, enforced by the browser before any tool runs | No equivalent documented |
| A record of what it did | Every step on the run’s timeline in the panel | An optional operation audit, kept on the daemon’s host |
| Telemetry | None | None |
| Interface languages | English | Ten, including Simplified and Traditional Chinese, Japanese and Korean |
Browsentic only
Type or speak a job in the side panel beside your tab and watch each step on its timeline, without switching to a terminal. BrowserSkill’s extension has no chat; every job starts from your agent.
One Chrome Web Store listing for every Chromium browser, Edge Add-ons for Edge, and a Mozilla-signed Firefox add-on that updates itself. BrowserSkill supports Chrome and Edge.
Talk to the page with the panel put away, or hold a key to talk while it works, wherever the browser supports speech.
A scheduled task runs an instruction once, on chosen days or at an interval, in a background tab. A monitor watches a page and tells you when something changes, with no agent polling it.
Submitting a form, uploading a file or leaving the run’s sites waits for you in the panel. Passwords reach the agent as placeholders it can type but never read.
Show it a job once and replay it later, let it map a site into a skill, or let it write a live tool you approve. It also solves “I am not a robot” checks on the way.
BrowserSkill only
Some of these are good reasons to choose BrowserSkill.
BrowserSkill works in a separate Agent Window, so you keep using your own tabs while it runs, and it borrows one of yours only after you confirm. Browsentic works in the tab beside the panel.
It ties each action to its requests, responses and console output, flags slow and duplicate API calls, mocks, blocks or replays requests, and exports the evidence. Browsentic reads network and console output, with no mocking or replay.
An agent running on a server can pair with the browser on your computer over authenticated WSS, with no inbound port. Browsentic binds to 127.0.0.1 only.
bsk is a single Rust binary that updates itself and rolls back if a new version fails to start. Browsentic Bridge needs Node.js 20 or newer, which the Mac and Windows apps install for you.
It records a tab as a silent MP4, exports tables and lists as JSON or CSV, and can keep an operation audit on the daemon’s host.
Its extension is translated into ten languages, including Chinese, Japanese and Korean, and it supports agents such as CodeBuddy, WorkBuddy and DeepSeek Harness. Browsentic is in English only.
The short version
Pick BrowserSkill if your work starts in an agent’s terminal and you want the browser kept in a window of its own, if you debug websites through the agent, or if the agent runs on a server.
Pick Browsentic if you want to ask from the browser itself, by text or voice, in Firefox as well as Chromium browsers, with approvals, schedules and recordings, and MCP for when you are in a terminal.
Both are free and open source, both run on an agent you already have rather than an API key, and both keep your logins on your computer with no telemetry.
Questions
BrowserSkill is an open-source project from Tencent, first published on GitHub in June 2026 under the MIT license. It pairs a command line called bsk, which includes a daemon, with an extension for Chrome and Edge, and installs a skill into agents such as Claude Code, Codex or Cursor. The agent then uses your logged-in browser in a separate Agent Window.
BrowserSkill is used from your agent: you ask in its terminal, and it works in a window of its own. Browsentic is used from the browser: you ask in a side panel beside your tab, typed or spoken, and Browsentic Bridge starts the agent CLI you already have. Browsentic also runs in Firefox, and adds scheduled tasks, monitors, recordings, approvals and sealed passwords. BrowserSkill goes further on website debugging and can serve an agent running on another machine.
BrowserSkill supports Chrome and Edge, and says other Chromium browsers may work without support. Browsentic runs in Firefox through a Mozilla-signed add-on that updates itself. Nine tools that depend on Chrome’s DevTools Protocol are missing there; everything else works as in Chrome, Edge, Brave or Arc.
Yes. Browsentic Bridge also serves MCP, so Claude Code, Cursor, Zed or any other MCP client can drive the same browser from a terminal or an editor. The side panel stays available for jobs you would rather ask for in the browser.
No. BrowserSkill is used by the agent you already run, and Browsentic starts the agent CLI you already signed in to, such as Claude Code, Codex, Antigravity or Cursor CLI. Neither one has a per-token bill of its own.
Yes. It is Apache 2.0 licensed, free, and needs no account and no subscription of its own. The source is on GitHub, and Browsentic Bridge runs on your computer, so there is no hosted service that could be switched off.
Checked against the BrowserSkill repository, its bundled skill and its store listings, 14 October 2026.
Anthropic’s agent in a Chrome side panel, compared with the agent CLI you already run.
An agent on your own API key in Chrome and Edge, compared with the agent CLI you already have.
One command, a pairing code, and the browser you already have open.
Free and open source under Apache 2.0, with no API key, subscription or account. One line installs Browsentic Bridge and opens the extension’s store page. Nothing connects until you pair the two.
Available in the Chrome Web Store
Available on Edge Add-ons
Firefox add-on, signed by Mozilla