Product overview
What is Mirascope?
Mirascope is a Python toolkit for AI engineers building applications and agents with large language models. The homepage demonstrates declaring a Python function as an LLM tool, defining a model call, and running an agent loop that executes tools and resumes the model with their outputs.
The same example applies automatic versioning, tracing, and cost tracking through Mirascope Ops. Traces expose versions, timing, input and output token counts, costs, messages, model thoughts, tool calls, tool inputs, and tool outputs so developers can inspect and iterate on application behavior. The example lists OpenAI, Anthropic, and Google as model providers.
Core Features
- Python model calls: Defines LLM calls using Python functions and decorators.
- Tool support: Turns Python functions into tools that models can invoke.
- Agent loops: Executes requested tools and resumes the model with results.
- Automatic versioning: Tracks changes to LLM application behavior.
- Tracing: Records messages, reasoning data, tool calls, inputs, and outputs.
- Cost tracking: Displays token counts and cost for individual traces.
Use Cases
- Tool-using agents: Developers can connect model decisions to typed Python functions.
- LLM application debugging: Engineers can inspect the sequence of messages and tool calls behind an output.
- Prompt iteration: Teams can compare versions, timing, tokens, and costs while refining behavior.
- Multi-provider development: Applications can work with models from several providers through one toolkit.
Frequently Asked Questions
What language is shown for Mirascope development?
The homepage demonstrates a Python API based on decorators and ordinary functions.
What information appears in traces?
The example shows versions, timing, input and output tokens, costs, messages, thoughts, and tool execution details.
