BrowserBC - an open-source browser action tracking generation skill from Einsia AI.

What is BrowserBC?

BrowserBC is an open-source project released by Navers Lab under Einsia AI. Its core goal is to transform human actions in a browser into reusable natural language skills, allowing Web Agents to "clone" human operational experience and complete similar tasks more efficiently. Users only need to perform the task once in the browser, and the system can automatically distill this operation into a structured "skill card" that can be given to any model for execution, achieving "do it once, reuse it countless times".

BrowserBC's main functions

• Track recordingIt records the entire task process through browser extensions, including page screenshots, DOM snapshots, user actions (clicks, inputs, redirects), page feedback, and the final state.
• Intelligent distillationAfter cleaning and denoising the original noisy trajectory, it is semantically segmented into coherent sub-processes and then transcribed into structured natural language skill cards. Fragile details such as coordinates and selectors are removed, while transferable procedural knowledge such as "what to do, how to judge completion, and what to do if it fails" is retained.
• Skill Map ManagementOrganize massive amounts of skills into a scalable skill graph, automatically merging, specializing, and linking related skills through semantic similarity to avoid duplication and redundancy.
• Skills retrieval and executionThe agent autonomously executes actions based on the retrieved skill card guidance and real-time page status, rather than mechanically replicating coordinates.
• Local deploymentIt provides a Journey Forge Local version, runs purely in Python, stores all data locally, and supports integration with Claude Code and Claude Desktop.

Follow us on WeChat and reply with "open source",join inAI open source project discussion group

How to use BrowserBC

• Configure and run the local serverCopy the configuration file and set the LLM API Key. After starting the server, access the control panel at http://127.0.0.1:8099/.
• Build and load browser extensions:exist extension Install the dependencies and build in the directory, then load it in developer mode in Chrome extensions manager. extension/dist/chrome-mv3.
• Recording taskUse the extension to record browser operations, mark the task intent after stopping, and upload.
• Automatic distillationThe background pipeline automatically completes the atomize → classify → bucket → distill → install process, generating skills in approximately 1-3 minutes.
• Use skills:
  • Claude CodeSkills are automatically installed to ~/.claude/skills/ In the directory.
  • Claude DesktopDownload from Control Panel .zip Upload the file in Settings → Skills.
• Configure browser execution(Optional): Configure Playwright MCP in the Control Panel to enable Claude Desktop to actually click/type/navigate.

BrowserBC's core advantages

• Skills are transferableSkills are "procedural priors" in the form of natural language, which can be freely transferred and reused between different models. A strong model can be distilled once, and a small model can also be reused cheaply.
• Stripping away fragile detailsInstead of cloning coordinates or DOM selectors, it clones "what to do + how to determine completion", so that the skills can still be applied to different situations when the page layout changes.
• Significantly improve success rateSuccess rate improved from 60.5% to 81.4% (+20.9%) on WebArena-Hard and from 32.9% to 68.4% (+35.5%) on ClawBench.
• Improve interaction efficiencyThe average number of tool calls decreased from 31.2 to 22.7 (a reduction of 27.3%), effectively reducing trial-and-error navigation.
• Cross-domain migration potentialProcedural priors can cross browser boundaries, and preliminary studies show that they can be migrated to the OSWorld desktop environment.
• Data privacy and securityThe local deployment version stores all data locally and only calls the configured LLM API during distillation.

BrowserBC's official website address

• Project Bloghttps://lab.einsia.ai/browserbc/
• GitHub repositoryhttps://github.com/Einsia/Browser-BC
• paperhttps://lab.einsia.ai/browserbc/paper

BrowserBC's Competitive Comparison

Application scenarios of BrowserBC

• Travel booking and price comparisonEnter the dates, location, and number of people on booking websites such as Airbnb and Booking. Apply filters (ratings, prices, facilities), read the search results, and sort them to find the best accommodation options.
• E-commerce shopping and filteringThe process involves searching for products on a shopping platform, applying multiple filtering criteria (price range, brand, rating), comparing product details, adding items to the shopping cart, and completing the checkout process.
• Backend management operationsPerform repetitive management tasks such as product listing, order processing, inventory updates, and user permission configuration in the e-commerce backend, CMS system, or management panel.
• Community forum interactionSearch for topics, filter posts, participate in discussions, gather information, or perform moderator management tasks on forums such as Reddit.
• Cross-site information integrationTasks that require collecting, comparing, and integrating information from multiple websites, such as cross-platform price comparison, multi-source data aggregation, and competitor monitoring.
• Development tools usagePerform code repository operations (login, create projects, submit issues, merge requests), CI/CD configuration, or documentation viewing on development platforms such as GitHub and GitLab.