AgentCraft AI: An Open-Source Multi-Agent Workspace Powered by Open-Weight LLMs

This article is a submission for the Hacktoberfest Open-Source AI Challenge: Week 1.

🌟 Inspiration & Background

Building with AI shouldn't require sending private codebase architecture or sensitive developer documentation to proprietary cloud APIs.

With the rapid advancement of open-weight AI models (such as Google's Gemma 2, Meta's Llama 3, and local runtime tools like Ollama and Hugging Face Transformers), developers now have the power to run high-performance AI agents completely offline on local hardware.

I created AgentCraft AI — a lightweight, modular multi-agent framework designed to orchestrate specialized AI agents for code auditing, documentation synthesis, and project planning, all powered by open-weight AI.

✨ What It Does

AgentCraft AI provides three specialized developer agents:

🏗️ Code Architect Agent: Inspects code structure, identifies methods/classes, audits security risks, and suggests architectural refactoring.
📑 Docs Summarizer Agent: Chunks and synthesizes lengthy technical specifications or API docs into executive summaries and bullet points.
🎯 Task Planner Agent: Decomposes high-level developer goals into step-by-step implementation milestones and generates dynamic Mermaid workflow diagrams.

🏗️ Architecture & Multi-Agent Flow

flowchart TD User["Developer"] -->|Submits Code / Task Goal| AgentCraft["AgentCraft AI Workspace"] AgentCraft --> Architect["Code Architect Agent"] AgentCraft --> Summarizer["Docs Summarizer Agent"] AgentCraft --> Planner["Task Planner Agent"] Architect -->|Local Inference| OpenWeightLLM["Open-Weight LLM (Gemma 2 / Llama 3 / Ollama)"] Summarizer -->|Local Inference| OpenWeightLLM Planner -->|Local Inference| OpenWeightLLM OpenWeightLLM -->|Parsed Output| Results["Structured JSON & Mermaid Diagrams"] Results --> User

💻 Open-Source Code & Setup

The project is 100% open-source under the MIT License.

📂 GitHub Repository: samainakhatun115-cpu/agentcraft-ai
🚀 Tech Stack: Python 3.9+, Open-Weight Models (Gemma 2 / Ollama / Hugging Face), Standard JSON Parsers.

Running AgentCraft AI:

from app import AgentCraft # Initialize Workspace with an open-weight model engine framework = AgentCraft(model_name="Gemma-2B / Llama-3-OpenWeight") # 1. Run Code Architect Agent code_analysis = framework.architect_agent(code_snippet) # 2. Run Docs Summarizer Agent summary = framework.summarizer_agent(doc_text) # 3. Run Task Planner Agent plan = framework.planner_agent("Deploy Local AI Code Auditor")

🎯 Hacktoberfest 2026: AI Belongs to Everyone

Hacktoberfest 2026's central mission — "AI belongs to everyone" — highlights the crucial role of open-source models and toolings. By running multi-agent workflows with open-weight models, developers maintain complete privacy, transparency, and freedom from vendor lock-in.