What is MechaForge Lab?
MechaForge Lab is a visual Meta-Agent builder and local execution studio. It empowers developers and workflow engineers to translate natural language descriptions into fully functional, production-ready AI agent architectures — visualize their structures, and execute them safely inside a local sandbox using real-world integrations.
You describe your agent in plain English. MechaForge compiles, runs, and lets you download a complete standalone Python package.
The Problem It Solves
Building multi-agent systems today is painful:
🔧 Endless boilerplate — wiring tools, credentials, and orchestration by hand 🧩 Framework complexity — LangChain, AutoGen, ADK all require deep expertise 🔐 Credential hell — manually setting up OAuth, env vars, and API keys 🕳️ Prototype-to-execution gap — your visual idea never makes it to running code MechaForge Lab bridges all of this in a single studio: describe → visualize → compile → run → download.
User Prompt │ ▼ Next.js Web UI Studio │ ▼ Meta-Agent Orchestrator (Oxlo / DeepSeek / Groq / Gemini) │ ▼ AgentProjectConfig JSON ├──────────────────────────┐ ▼ ▼ Visual ReactFlow Graph Python Code Generator │ ▼ agent.py / runner.py │ ▼ Local Python Sandbox (.venv) │ ▼ Runner Process │ ┌────────────────┼────────────────┐ ▼ ▼ ▼ Google Sheets Google Drive DuckDuckGo / Wikipedia └────────────────┼────────────────┘ ▼ Active LLM (Groq / Oxlo / Gemini)
Pipeline Stages
Orchestration — The frontend sends the builder prompt to the Meta-Agent Orchestrator (powered by Oxlo/DeepSeek, Groq, or Gemini). Structural Visualizer — The LLM returns an AgentProjectConfig JSON block that ReactFlow parses to render an interactive node graph. Local Compilation — The code generator writes a self-contained Python agent bundle (agent.py, runner.py, requirements.txt, .env) to a local execution directory (.agent_run). Sandbox Execution — The backend runs the Python script inside a virtual environment (.venv), queries the selected LLM, and streams tool results and logs back to the web console.
Google ADK: Model-Agnostic Evolution
Originally built around the Google Agent Development Kit (ADK) — designed exclusively for Gemini — MechaForge Lab modifies this schema into a model-agnostic runtime.
Unified OpenAI-Compatible Client — A wrapper in the generated Python runtime binds different API signatures (Groq, Gemini, Oxlo) into standard OpenAI SDK completions. Flexible Config Mapping — System instructions, schemas, and completion parameters are dynamically normalized in the backend, allowing a Llama or DeepSeek model to plan and execute with the same accuracy as a native Gemini model.
Log in or sign up for Devpost to join the conversation.