Pythinker

<div align="center">

<img src="frontend/public/icon.svg" alt="Pythinker" width="120" />

# Pythinker

### 🚀 The Open-Source AI Agent That Does It All

**Browse the web. Write & run code. Research anything. Generate beautiful reports.**

*Your self-hosted alternative to [Manus AI](https://manus.im) — with 43+ tools and full autonomy.*

[![CI](https://github.com/mohamed-elkholy95/Pythinker/actions/workflows/test-and-lint.yml/badge.svg?branch=main-dev)](https://github.com/mohamed-elkholy95/Pythinker/actions/workflows/test-and-lint.yml)
[![Docker](https://github.com/mohamed-elkholy95/Pythinker/actions/workflows/docker-build-and-push.yml/badge.svg)](https://github.com/mohamed-elkholy95/Pythinker/actions/workflows/docker-build-and-push.yml)
[![Security](https://github.com/mohamed-elkholy95/Pythinker/actions/workflows/security-scan.yml/badge.svg)](https://github.com/mohamed-elkholy95/Pythinker/actions/workflows/security-scan.yml)
[![GitHub issues](https://img.shields.io/github/issues/mohamed-elkholy95/Pythinker)](https://github.com/mohamed-elkholy95/Pythinker/issues)
[![GitHub stars](https://img.shields.io/github/stars/mohamed-elkholy95/Pythinker)](https://github.com/mohamed-elkholy95/Pythinker/stargazers)
[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
[![Python 3.12+](https://img.shields.io/badge/Python-3.12+-3776AB.svg)](https://python.org)
[![Vue 3](https://img.shields.io/badge/Vue-3-4FC08D.svg)](https://vuejs.org)
[![FastAPI](https://img.shields.io/badge/FastAPI-009688.svg)](https://fastapi.tiangolo.com)
[![Docker](https://img.shields.io/badge/Docker-2496ED.svg)](https://docker.com)
[![Scrapling](https://img.shields.io/badge/Scraping-Scrapling-blue)](https://github.com/D4Vinci/Scrapling)

[🌐 Website](https://pythinker.com) · [📖 Docs](https://github.com/mohamed-elkholy95/Pythinker/wiki) · [🐛 Report Bug](https://github.com/mohamed-elkholy95/Pythinker/issues) · [💡 Request Feature](https://github.com/mohamed-elkholy95/Pythinker/issues) · [💬 Discussions](https://github.com/mohamed-elkholy95/Pythinker/discussions)

</div>

---

## ⭐ Give us a star!

If you find Pythinker useful, please consider giving us a **star** ⭐ on GitHub — it helps others discover this project and motivates continued development!

---

## 🎯 What is Pythinker?

Pythinker is an **open-source, self-hosted AI agent platform** that can autonomously browse the web, write & execute code, search the internet, manage files, and deliver polished research reports — all from a beautiful real-time interface.

> **Think of it as your personal AI researcher + developer + assistant — running entirely on your own infrastructure.**

Built with ideas from [Manus AI](https://manus.im), [OpenManus](https://github.com/manusai/OpenManus), and [Nanobot](https://github.com/ArcadeAI/nanobot).

**Author:** [Mohamed Elkholy](https://github.com/mohamed-elkholy95)

---

## 🔥 Why Pythinker?

| | Pythinker | Manus AI | ChatGPT | Other OSS Agents |
|---|:---:|:---:|:---:|:---:|
| **Open Source** | ✅ | ❌ | ❌ | ✅ |
| **Self-Hosted** | ✅ | ❌ | ❌ | ✅ |
| **43+ Built-in Tools** | ✅ | ✅ | ❌ | ❌ |
| **Live Browser Streaming** | ✅ | ✅ | ❌ | ❌ |
| **Multi-LLM Support** | ✅ | ❌ | ❌ | ⚠️ |
| **Report Generation** | ✅ | ✅ | ❌ | ❌ |
| **Telegram Bot** | ✅ | ❌ | ❌ | ❌ |
| **Sandboxed Execution** | ✅ | ✅ | ✅ | ⚠️ |
| **Hybrid Memory (RAG)** | ✅ | ❌ | ✅ | ❌ |
| **MCP Tool Integration** | ✅ | ❌ | ❌ | ⚠️ |
| **Free** | ✅ | ❌ | ❌ | ✅ |

---

## ✨ Key Features

| Feature | Description |
|---------|-------------|
| 🛠️ **43+ Built-in Tools** | File, browser, shell, search, code, messaging, automation — the agent picks the right tool for every step |
| 🌐 **Live Browser Streaming** | Watch the agent browse in real-time via CDP screencast — take over control at any moment |
| 📊 **Beautiful Report Generation** | Automatically produces structured, citation-rich research reports with charts and references |
| 🔒 **Sandboxed Execution** | Every task runs in an isolated Docker container with Chrome, Python, Node.js, and shell access |
| 🤖 **Multi-Model Support** | Works with any OpenAI-compatible API — GPT-4o, Claude, DeepSeek, Kimi, GLM, local models, and more |
| 🕷️ **Intelligent Web Scraping** | Three-tier scraping with automatic escalation — HTTP, dynamic rendering, stealth browser (powered by [Scrapling](https://github.com/D4Vinci/Scrapling)) |
| 📱 **Telegram Integration** | Full-featured Telegram bot gateway with inline buttons, file sharing, and streaming responses |
| 🔌 **MCP Tool Integration** | Extend capabilities with external Model Context Protocol servers |
| 🧠 **PlanAct Agent Architecture** | Intelligent planning, execution, reflection, and verification pipeline with 108 agent modules |
| 💾 **Hybrid Memory System** | Semantic search over past sessions via Qdrant (BM25 + dense retrieval with RRF fusion) |

---

## 🏗️ Architecture

```
+-----------------------------------------------------------------------+
|                         PYTHINKER PLATFORM                            |
|                                                                       |
|  +----------+  +---------------------------------------------------+  |
|  | Frontend  |  |              Backend (FastAPI)                    |  |
|  | Vue 3     |<>|  PlanAct Agent . 43+ Tools . SSE Streaming       |  |
|  | TypeScript|  |  Model Router  . DDD Services . Report Gen       |  |
|  +----------+  +--------------+-------------------+----------------+  |
|                               |                   |                   |
|  +----------+  +--------------v--+  +------------v--------------+    |
|  | Telegram  |  |    Sandbox(es)  |  |       Data Layer          |    |
|  | Gateway   |  |  Ubuntu Docker  |  | MongoDB . Redis . Qdrant  |    |
|  | (Nanobot) |  |  Chrome . Python|  | MinIO (Object Storage)    |    |
|  +----------+  +-----------------+  +---------------------------+    |
+-----------------------------------------------------------------------+
```

---

## 🛠️ Tech Stack

| Layer | Technology |
|-------|------------|
| **Frontend** | Vue 3, TypeScript, Vite, TipTap, Monaco Editor, xterm.js, Plotly |
| **Backend** | FastAPI, Python 3.12+, Pydantic v2, Beanie ODM, SSE, WebSockets |
| **Agent** | PlanAct pipeline, adaptive model routing, tool efficiency monitoring, reflection & verification |
| **Web Scraping** | [Scrapling](https://github.com/D4Vinci/Scrapling) (tiered: HTTP/TLS impersonation → dynamic rendering → stealth browser), Playwright |
| **Sandbox** | Ubuntu Docker, Chromium, Playwright, Supervisord, CDP screencast |
| **Messaging** | Nanobot multi-channel gateway (Telegram, Slack, Discord, DingTalk, Feishu) |
| **Database** | MongoDB 7.0 (sessions & state), Redis 8 (cache & coordination) |
| **Vector Search** | Qdrant (semantic memory, hybrid BM25 + dense retrieval) |
| **Object Storage** | MinIO (file uploads, artifacts, report assets) |
| **CI/CD** | GitHub Actions (lint, test, security scan, Docker build) |
| **Monitoring** | Prometheus, Grafana, Loki, Promtail |

---

## 🧰 Tool Categories

The agent has access to **43+ tools** organized into 10 categories:

| Category | Tools | What They Do |
|----------|-------|-------------|
| 📁 **File** | read, write, list, search, upload, download | Full filesystem access within the sandbox |
| 🌐 **Browser** | navigate, click, type, screenshot, scroll, evaluate JS | Headless Chrome with live CDP streaming |
| 🔍 **Search** | web search, scrape, extract | Internet research with multiple providers (Tavily, Serper, Exa) |
| 💻 **Shell** | execute, background, interactive | Full terminal access with real-time output |
| 💬 **Message** | ask user, notify, report | Communication and deliverable generation |
| 🔌 **MCP** | external tool servers | Extensible via Model Context Protocol |
| 🧬 **Code** | analyze, refactor, test | Code intelligence and manipulation |
| 📋 **Plan** | create plan, update step, checkpoint | Structured tas

[truncated…]