Architecture
Learn how Agentsmith works
At its core, Agentsmith is a platform designed to bridge the gap between prompt engineering and production-ready code. It provides a structured, version-controlled, and type-safe workflow for developing and deploying AI agents.
Agentsmith consists of several key components that work together to provide a seamless experience from authoring to execution.
- Agentsmith Studio: A web-based interface for authoring, testing, and managing your prompts. This is where non-technical team members can collaborate with developers.
- GitHub: Your Git repository acts as the single source of truth for all versioned prompts.
- Your Application: The codebase where you install the Agentsmith SDK to consume and execute prompts.
- OpenRouter: The gateway we use for all LLM calls. This provides flexibility, allowing you to switch between models from different providers without changing your code.
We use OpenRouter to provide a unified API for a vast range of LLMs. This means you can experiment with different models, find the most cost-effective one for your needs, and switch seamlessly, all without being locked into a single provider.
The Development Workflow
The core philosophy of Agentsmith revolves around a "Git-as-source-of-truth" workflow that combines the best of visual editing and code-based reliability.
- Author & Test: Prompts are created and tested in the visual Studio.
- Sync & Version: Prompts are synced to your Git repository as versioned files. This creates a durable, reviewable history of all changes.
- Develop & Execute: The type-safe SDK consumes these files in your application, ensuring that the prompts you execute match what's in your repository.
This workflow ensures that changes to prompts can be reviewed, tested, and deployed with the same rigor as any other code change in your project.
To learn more about the details, explore the following sections: