🚀 SourceBot

⚡ A Devin-style autonomous coding agent framework
Plan → Execute → Iterate with DAG-based workflows

---

⚡ TL;DR

SourceBot is an autonomous agent runtime that:

• Plans tasks as a DAG
• Executes them with tools + LLMs
• Maintains memory and state
• Supports retry, resume, and parallel execution

---

🔥 Why SourceBot?

Most AI tools today are chat-based.

SourceBot is different.

It is built to plan, execute, and iterate — like a real developer.

• 🧠 Plans tasks using a DAG (not linear prompts)
• 🔧 Uses tools (shell, skills, APIs)
• 🔁 Retries, resumes, and replays execution
• 💾 Maintains memory across steps
• 🐳 Runs safely in sandboxed environments

This is not a chatbot. This is an agent runtime.

---

🎬 Demo

sourcebot init
sourcebot init_workspace
sourcebot cli

⚡ Example: Build a Python HTTP server autonomously

⚡ Demo is shortened for clarity

---

🧠 What Makes It Different

🔀 DAG-Based Execution (Core Innovation)

Unlike traditional agents:

Prompt → LLM → Response

SourceBot:

Task
 ↓
Decomposition
 ↓
DAG Graph
 ↓
Parallel Execution
 ↓
Tool + LLM Calls
 ↓
State + Memory Update

✅ Parallelizable ✅ Retryable ✅ Resumable ✅ Observable

---

🧩 Core Features

🧠 Agent Runtime

• Modular architecture
• Identity + context system
• Multi-provider LLM abstraction

---

🔀 DAG Execution Engine

• Task decomposition
• Parallel scheduling
• Retry policies
• Run state persistence
• Partial replay

---

🛠 Tool System

• Tool registry
• Shell execution
• Skill-based extensions

---

💾 Memory System

• Sliding window memory
• LLM-based summarization
• Persistent storage

---

💬 CLI Interface

• Interactive agent shell
• Safe runner
• Workspace bootstrap

---

🐳 Sandbox Execution

• Docker isolation
• Safe tool execution

---

🏗 Architecture

🔷 High-Level Architecture

graph TD
    U[User] --> CLI[CLI Interface]
    CLI --> AR[Agent Runtime]
    AR --> DP[DAG Planner]
    DP --> TG[Task Graph]
    TG --> DS[DAG Scheduler]
    DS --> TE[Tool Executor]
    DS --> LLM[LLM Clients]
    TE --> SB[Docker Sandbox]
    LLM --> MEM[Memory System]
    SB --> MEM
    MEM --> OUT[Output]

Loading

---

🔷 Execution Flow

graph LR
    A[User Task] --> B[Task Decomposition]
    B --> C[DAG Generation]
    C --> D[Parallel Execution]
    D --> E[Tool Calls / LLM]
    E --> F[State Update]
    F --> G[Final Result]

Loading

---

📁 Project Structure

sourcebot/
├── bus/              # Event system
├── cli/              # CLI interface
├── config/           # Config system
├── context/          # Context + skills
├── conversation/     # Conversation layer
├── docker_sandbox/   # Execution isolation
├── llm/              # LLM abstraction
├── memory/           # Memory system
├── runtime/          # Agent + DAG engine
├── tools/            # Tooling system
├── session/          # Persistence
└── utils/            # Utilities

---

🚀 Installation

git clone https://github.com/DavidVi2Bot/sourcebot.git
cd sourcebot
pip install -e .

---

🧪 Development

pip install -r requirements-dev.txt
pytest
black .
ruff check .

---

🛣 Roadmap

🧱 Core

• DAG execution engine
• Tool system
• CLI runtime
• Memory system

🚧 Next

• Web UI (IDE-like)
• Multi-agent collaboration
• Plugin / marketplace system
• Long-term memory (vector DB)
• Remote execution / cloud runtime

---

🧠 Vision

SourceBot aims to become a general-purpose autonomous agent framework:

• 🧑‍💻 Coding agents (Devin-style)
• ⚙️ DevOps automation
• 🔬 Research agents
• 🔄 Workflow orchestration

---

⚠️ Status

🚧 Alpha — rapidly evolving

Expect breaking changes and fast iteration.

---

🤝 Contributing

We welcome contributions!

• Bug reports
• Feature ideas
• Tools / skills
• Documentation improvements

---

⭐ Star History

If this project interests you, give it a ⭐ — it helps a lot!

---

📄 License

MIT License

---