mirascope

Why we included this project

Mirascope fills the gap between raw provider SDKs and heavyweight agent frameworks. It wraps providers like Anthropic and OpenAI behind a small set of typed primitives (call, stream, resume, tool), so your code stays model-agnostic while you still manage each message, content block, and tool call yourself. The @llm.call and @llm.tool decorators, Pydantic-backed structured output, and a response.resume loop make building a custom agent straightforward without hiding what actually gets sent to the model. Teams that find LangChain too abstract but raw provider APIs too tedious tend to land here; because it stays a library rather than a framework, your own architecture and loop logic stay front and center. It is a good fit when you want portable LLM code that you still fully own and can reason about.