poppy

Poppy implements a schema-based function calling mechanism that allows users to define and invoke functions across multiple model providers seamlessly. This is achieved through a unified interface that abstracts the underlying API differences, enabling developers to switch between providers like OpenAI and Anthropic without changing their codebase. The architecture leverages a plugin system that dynamically loads provider-specific implementations based on user configuration, ensuring flexibility and extensibility.

Poppy features a contextual state management system that retains user session data across multiple interactions with AI models. This is implemented using a lightweight in-memory store that captures the context of previous calls and allows for stateful interactions. The architecture supports both ephemeral and persistent states, enabling developers to choose the appropriate context retention strategy based on their application needs.

Poppy supports dynamic API orchestration that allows developers to define complex workflows involving multiple AI models and external APIs. This capability is facilitated through a visual workflow editor that generates the necessary orchestration code based on user-defined steps. The system employs a microservices architecture that enables independent scaling of different workflow components, ensuring high availability and performance.