Mirascope – Open-Source Python LLM Anti-Framework, Type-Safe Tools & Agent Loops

#Quasa #QUA #Mirascope

Mirascope is an open-source Python library described as an LLM anti-framework. It gives type-safe, provider-agnostic tools for calling models from OpenAI, Anthropic, Google, and others with a simple decorator or model object. You define prompts as functions, add tools with @llm.tool, request structured output via Pydantic, stream responses, and build agent loops by repeatedly calling execute_tools and resume. It includes automatic versioning, tracing, and cost tracking on decorated calls. Local models and MCP integration are supported. There is no commercial SaaS pricing; you install it with uv add "mirascope[all]" or pip and bring your own API keys. The design prioritizes control and composability over opinionated agent frameworks.

𝐂𝗢𝗥𝗘 𝗦𝗧𝗥𝗘𝗡𝗚𝗧𝗛𝗦
• Provider-agnostic calls with one interface
• Type-safe tools and structured output
• Simple agent loops without a heavy framework
• Built-in tracing, versioning, and cost tracking
• Lightweight and composable  

𝗜𝗗𝗘𝗔𝗟 𝗙𝗢𝗥
Python developers who want production LLM features without locking into a full agent framework.

𝗛𝗜𝗚𝗛𝗟𝗜𝗚𝗛𝗧𝗦
• Decorator-based prompts and calls
• Streaming and async support
• Resume pattern for multi-turn tool use
• Local model and MCP options
• Open-source with clear examples  

𝗣𝗢𝗧𝗘𝗡𝗧𝗜𝗔𝗟 𝗖𝗢𝗡𝗦𝗜𝗗𝗘𝗥𝗔𝗧𝗜𝗢𝗡𝗦
• You manage your own model API keys and costs
• No hosted platform or visual builder
• Agent logic is explicit rather than abstracted
• Newer models require updates to the library
• Observability is basic compared with dedicated platforms  

𝗢𝗩𝗘𝗥𝗔𝗟𝗟 𝗩𝗘𝗥𝗗𝗜𝗖𝗧
4.3/5 stars. Use Mirascope when you want clean, typed LLM code that stays under your control. It is not a no-code agent builder. Earn 1 QUA reward by reviewing on Quasa.io too!

𝗚𝗘𝗧 𝗦𝗧𝗔𝗥𝗧𝗘𝗗: https://quasa.io/projects/mirascope