AI browser automation inside Chrome with explicit tool calls, supervised execution, and connector-ready outputs.

• Browser agent for real workflows
• Explicit tools instead of blind scripts
• Plan approval before sensitive execution
• Background runs with session recovery
• Connectors for downstream delivery

Scenario: Open Wikipedia and find the article about Albert Einstein.

BrowseAgent is a Chrome Manifest V3 extension that turns plain-English goals into browser actions. It can navigate pages, read the DOM, click through flows, extract structured data, call external APIs, and send results to connected tools. It is built for operators, researchers, growth teams, and engineers who need flexible browser automation without leaving Chrome.

Most browser automation sits in one of two weak categories: brittle scripts or opaque AI agents. Scripts fail when a page changes. Black-box agents hide how they decided to act and make debugging painful. BrowseAgent was built to keep LLM flexibility while forcing execution through explicit, auditable tools, visible run state, and Chrome-native boundaries.

• Navigate pages, tabs, history, and frames
• Read page text and accessibility-grounded structure
• Click, type, scroll, hover, select, and submit
• Extract structured data from repeated page patterns
• Call external APIs and webhooks during a run
• Route results to Slack, Notion, Airtable, Discord, Telegram, Sheets, email, or custom webhooks
• Preview execution with Plan mode before acting
• Run background jobs with chrome.alarms
• Recover unfinished sessions after panel restart
• Enforce blocklists, confirmations, and anti-loop limits

Full tool catalog -> docs/TOOLS.md

Components:

• src/background/ - service worker orchestration, alarms, recovery, routing
• src/content/ - DOM reading and page interaction in browser context
• src/agent/ - planning, reflection, state, and completion logic
• src/providers/ - interchangeable OpenAI-compatible LLM backends
• src/sidepanel/ - task entry, status, approvals, history, settings
• src/integrations/ - delivery adapters for external destinations

Flow:

User goal -> Side panel -> Service worker -> Agent loop -> Content script / provider / connector -> Result

graph LR
U["User Goal"] --> P["Side Panel"]
P --> SW["Service Worker"]
SW --> A["Agent Loop"]
A --> CS["DOM Execution"]
A --> LLM["LLM Provider"]
A --> I["Connector or API"]
CS --> A
LLM --> A
I --> A
A --> P

Architecture detail -> docs/ARCHITECTURE.md

• JavaScript ES modules
• Chrome Extensions Manifest V3
• Chrome Side Panel API
• chrome.tabs, chrome.scripting, chrome.alarms, chrome.storage, chrome.declarativeNetRequest
• OpenAI-compatible provider interface
• Node built-in test runner

1. Open chrome://extensions/ in Chrome 114 or newer.
2. Enable Developer mode and load browseagent-ext/ as an unpacked extension.
3. Open the side panel and configure a provider in Settings.
4. Click Test to verify the model connection.
5. Enter a goal in Task and run it.

Full provider setup -> docs/PROVIDERS.md

Run a simple browser research task:

Open Wikipedia and find the article about Albert Einstein.

Run a connector-oriented workflow:

Collect dashboard values from the current tab and send the summary to Slack.

Run the automated test suite:

npm test

src/         agent loop, service worker, content scripts, providers, UI, tools
docs/        architecture, providers, tools, release checklists
tests/       unit, integration, and e2e regression coverage
release/     packaged builds and release bundle contents

• Stage: Public Beta
• Version: 1.0.3
• Planned: broader provider coverage, workflow templates, better long-run observability

npm test

• Open an issue with the user problem or workflow gap
• Fork the repository and create a focused branch
• Add or update tests when behavior changes
• Submit a pull request with a clear before/after explanation

MIT - see LICENSE

If you like this project, please give it a star ⭐

For questions, feedback, or support, reach out to:

LinkedIn Email