Case study

Mogbot

Mogbot is a self-hosted Manus-style agent. A React UI talks over WebSocket + REST to a FastAPI backend that wraps the Anthropic API, a Playwright browser, and a code execution sandbox. The agent pauses for the user on CAPTCHAs, 2FA, or login walls instead of failing silently.

RoleAI engineering, full-stack architecture, browser automation

TimelinePersonal project

FocusAutonomous AI Agent

AI agent designBrowser automationCode execution sandboxSpend + history tracking

FastAPI + React + PlaywrightStack

Anthropic ClaudeModel

https://github.com/MaxwellM34/MogbotRepository

The brief

Challenge

Hosted agent platforms hide cost, can't be paused on auth challenges, and don't give you the full execution history.

Approach

What we made

Built a self-hosted control plane with explicit budget tracking, a Playwright browser the user can take over, and PostgreSQL persistence so every run is inspectable.

WebSocket + REST architecture between React UI and FastAPI control plane.
Playwright integration handles real browser interactions: click, type, scroll, fill forms.
Real-time CAD budget enforcement to keep autonomous runs from over-spending.
PostgreSQL-backed task history and logs for replay and auditing.
Pauses for human input on CAPTCHAs, 2FA, and login walls.

Outcome

Results

A working autonomous agent that completes browse + code + file workflows end-to-end while staying within a CAD budget and yielding control on human-only steps.

PythonFastAPITypeScriptReactPlaywrightPostgreSQLAnthropic APIDocker

Gallery

Visual snapshots

Click any image to expand.

Next project

Boot.dev Agent Project

Boot.dev coursework building a small Python agent against the Gemini API with a simple tool-calling loop and a calculator example.

View next project