Agentok Studio provides a visual, drag-and-drop interface for building and managing multi-agent applications powered by the AG2 framework. It targets developers and researchers looking to simplify the creation of complex agentic workflows, offering a diagram-based code generation approach for native Python execution.

How It Works

Agentok Studio visualizes AG2's agent interactions, allowing users to define relationships and tool integrations through a graphical interface. It generates self-contained Python code that relies solely on the official AG2 library, promoting transparency and ease of debugging. The system emphasizes clear data flow visibility and direct access to execution logs.

Quick Start & Requirements

• Online Demo: Visit https://studio.agentok.ai (not for production).
• Local (Docker): Requires Docker. Clone repo, copy .env.sample files, configure Supabase keys, and run docker-compose up -d.
• Local (No Docker): Requires Node.js (pnpm/yarn) for frontend, Python 3.x with Poetry for backend. AG2's latest version defaults to Docker for code execution; set AUTOGEN_USE_DOCKER=False in api/.env to disable. REPLICATE_API_TOKEN is needed for LLaVa.

Highlighted Details

• Diagram-based code generator for AG2.
• Full visibility into data representation and execution logs.
• Tool editor for creating and managing configurable tools.
• Supports light and dark themes.

Maintenance & Community

• Active development, with recent additions like a DeepSeek Assistant Node.
• Community engagement encouraged via GitHub Issues and Discord.
• Uses semantic-release for versioning and Commitlint for commit messages.

Licensing & Compatibility

• Licensed under Apache 2.0 with additional terms.
• Generated code is native Python, compatible with standard execution environments.

Limitations & Caveats

The online demo is not production-ready and may experience data wipes. The project is actively developing a new backend (Roboco) as an alternative to the current AG2 backend. RAG functionality has been removed.