AI-Native Browser Cluster: Orchestrate N Isolated Fingerprint Browsers with a Single Natural-Language Command
An open-source project that combines 「fingerprint browsers」 and 「AI Agent」 into one: Matrix Agent—no more manually configuring a dozen browser environments or writing piles of scripts to operate pages. Just tell it what to do in natural language, and it handles the orchestration, execution, and audit trail itself.
What problems does it solve?
Multi-account matrix operations (e-commerce / social media / data collection) face three major challenges:
- Environment isolation: Each account needs an independent browser environment + fingerprint + proxy
- Repetitive work: Logging in, checking, and batch operations are all mechanical tasks
- Uncontrollable processes: When something goes wrong, you don’t know which step failed; getting stuck on a CAPTCHA brings everything to a halt
Matrix Agent solves all three together: the fingerprint browser cluster manages environments, the AI Agent handles execution, and step-by-step audit trails + recordings enable review.
Core capabilities
- AI task execution: Natural-language instruction → observation → decision → execution loop, supporting any OpenAI-compatible API (Base URL / key / model are all configurable)
- Fingerprint browser cluster: Isolated multi-profile environments, with an independent userDataDir / fingerprint / proxy for each profile; grouping, cloning, and import/export (including cookies)
- Batch tasks: Send one instruction to multiple profiles for parallel execution, with results aggregated automatically; tasks can include a profile pool and switch profiles as needed during execution
- Process recording and playback: After the AI successfully explores a workflow, it automatically records the action sequence; playback does not call the LLM at all. When XPath fails, it self-heals through fuzzy matching by tag + text, and only hands control back to the LLM if self-healing fails
- Human–machine collaboration: Automatically pauses and opens a dialog when a CAPTCHA / human verification appears (screenshot + AI reasoning + recent actions), then resumes after a person handles it
- Scheduled tasks: Two rule types—interval / daily—with multiple profiles automatically running batch tasks
- Notifications: Desktop notifications + Webhooks (DingTalk / WeCom / self-hosted)
- Audit trails: Screenshots + reasoning process + complete execution recordings (webm) for every step; task details can be replayed
Anti-detection (for standard e-commerce / social media scenarios)
- Stealth patches: Remove webdriver, disable automation flags, preserve permissions / plugins / fonts, and set the Accept-Language q value
- Human-like behavior: Bezier mouse trajectories, randomized per-character typing rhythms, and segmented randomized scrolling
- Fingerprint consistency: Constrain the number of CPU cores based on GPU tier and inject system font lists according to the OS
- Proxy egress fingerprint matching: After entering a proxy, generate a matching time zone / Locale / language from the egress IP with one click
- WebRTC leak detection: Check whether ICE candidates bypass the proxy and expose the real IP
Scope note: This solution targets standard e-commerce / social media detection and does not promise to bypass strong detection systems such as CreepJS / FingerprintJS Pro. Countermeasure-grade fingerprinting requires source-level engine modifications and is planned for a later stage.
Tech stack
| Layer | Technology |
|---|---|
| Desktop framework | Electron 36 (single EXE installer) |
| UI | React 18 + Tailwind CSS + Vite 7 |
| Browser automation | playwright-core (system Chrome preferred) |
| Storage | better-sqlite3 (WAL + schema migrations) |
| Keys | Electron safeStorage encryption |
If you’re also working on matrix operations / automation, feel free to submit an Issue / PR. If you find it useful, give it a Star 
If anyone skilled in UI design is interested, feel free to improve my UI as well (I made it a bit ugly).
Avoid Getting Sanctioned
If anyone has the expertise, you could try integrating with platforms that use human verification; making a registration bot would also be possible ()
