🚀 AgentFlow | AI Agent Orchestration Platform

AgentFlow is a professional-grade, autonomous multi-agent orchestration platform. It empowers teams to build, monitor, and deploy specialized AI agents equipped with real-world tools. Engineered for high-performance reasoning and real-time telemetry, AgentFlow bridges the gap between complex LLM workflows and production-ready interfaces.

---

🏛️ System Architecture

AgentFlow utilizes a decoupled, event-driven architecture to ensure scalability and separation of concerns.

graph TD
    subgraph Client_Layer [Client Layer]
        UI[React 19 Dashboard]
        TG[Telegram Bot]
    end

    subgraph Backend_Layer [Orchestration Layer]
        API[FastAPI Server]
        LG[LangGraph Runtime]
        WS[WebSocket Telemetry]
    end

    subgraph Logic_Layer [Agent Intelligence]
        LLM[Gemini 3.1 Flash]
        Tools[Toolbelt: Search, Scraping, YT, Calc]
        Mem[Long-Term Memory Engine]
    end

    UI <--> WS
    TG <--> API
    API --> LG
    LG --> LLM
    LG --> Tools
    LG --> Mem

Loading

🎯 Key Engineering Decisions

• Orchestration: LangGraph: We use a State-Machine approach to allow for cycles, conditional routing, and persistent AgentState.
• LLM: Google Gemini: Selected for its 1M+ token context window and native multimodal capabilities (text + image).
• Static Asset Serving: To optimize performance, generated images are stored locally and served via a lightweight Markdown pointer, reducing WebSocket payload size.
• Async-First: The entire backend is built on FastAPI with async/await to handle concurrent LLM and Tool I/O without blocking.

---

Demo Video

Agent.Flow.Demo.mp4

---

📂 Project Structure Map

Directory	Purpose
agentflow-platform/backend/	Core FastAPI logic, API routers, and database models.
agentflow-platform/backend/runtime/	The Brain. Contains LangGraph logic, tool definitions, and LLM wrappers.
agentflow-platform/frontend/	React 19 source code, dashboard layouts, and ReactFlow canvas logic.
agentflow-platform/docs/	Deep-dive documentation for each phase of development.
start_platform.py	Single-command entry point to launch the entire ecosystem.

---

🛠️ Team Collaboration Guide

1. Adding a New Tool

To equip agents with new capabilities:

1. Open agentflow-platform/backend/runtime/tools.py.
2. Define your tool using the @tool decorator.
3. Add the function name to the AVAILABLE_TOOLS dictionary.
4. The tool will automatically be selectable in the Dashboard UI.

2. Modifying Workflow Logic

Workflow routing is handled in agentflow-platform/backend/runtime/graph_builder.py. You can adjust how agents hand off tasks or introduce new conditional nodes here.

---

🔐 Security & Setup

Local Installation

1. Environment: Copy .env.example to .env.
2. Keys: Add your GEMINI_API_KEY and TELEGRAM_BOT_TOKEN.
3. Launch: Run python start_platform.py.

---

📚 Technical Documentation

For deep-dives into specific modules, refer to our internal technical docs:

• 🚀 Architecture Deep Dive
• 🏗️ Backend Foundation
• 🧠 Agent Runtime
• 📊 Workflow Builder
• 🔄 Workflows, Agents & Tokenomics

---

📱 External Integration

The Telegram bot acts as a mobile extension of the platform. It uses the same workflow engine, ensuring that logic updated on the web dashboard is immediately available on mobile.

2. Define the nodes (agents) and edges (logic paths).
3. Run the script to inject the template into the agentflow_platform.db.