ClinkClang is an open-source TypeScript framework for building AI agent workflows with a focus on the developer’s experience and enterprise scalability.

Vision

(TODO)

Core Features

• Flexible Provider System: Easily switch between different LLM providers (OpenAI, Claude, DeepSeek).
• Strategy Based: Support for different reasoning strategies like ReAct and Reflexion.
• Tool Integration: Easy to use interface for adding custom tools and capabilities.
• Memory Management: (TODO).
• Structured Output: Schema validation and typed responses using the ZOD library.
• Built-in Evaluation : AutoEvals library integration for quality, safety, and task-specific metrics.

How It Works

The system is built around a modular core that manages:

User Input → Agent → Framework → Provider → Model → Framework → Tool Execution → Final Output

Each component:

• User : Input queries and tasks.
• Agent : Core orchestrator managing flow.
• Provider/Model : LLM integration (OpenAI, Claude, etc).
• Strategy : ReAct/Reflexion reasoning.
• Tools : Custom function execution.
• Output : Structured, validated responses.

Future Roadmap

(TODO)

Project Status (TODO)

ClinkClang is currently in active development.