Start here
Apizr helps you understand existing Python work, keep evidence from trusted executions, and share selected functions with applications or tool clients. Start with the result you need. Exposing an existing function requires no Experiment Run.
Choose your first result
| Your starting point | First action | Observable result |
|---|---|---|
| A script or notebook you want to understand | Inspect an experiment | Data references, parameters and unknowns, without running it |
| A function another application needs | Make your first calls | A real HTTP response or MCP tool result |
| Two training attempts | Compare fraud experiments | Recorded differences in inputs and results |
| A service ready for a colleague | Hand over a service | Checked runtime, bundle and delivery evidence |
Install for the action you chose
Use Python 3.11–3.14 on a qualified macOS or Linux environment. Install Apizr is the canonical installation procedure. The core is sufficient for static Python inspection and generation. The notebook extra enables conversion; scientific libraries belong to the workload environment. Generated services install their own requirements.
For a zero-execution first result, save example.py:
learning_rate = 0.05
Then, in the environment where you installed Apizr:
apizr --version
apizr experiment inspect example.py
Check that Parameters contains learning_rate = 0.05. The eight report sections
separate available evidence from unknown facts. No model has run and no metric
value has been measured. You can inspect a pandas script without installing
pandas; you need pandas when you choose to execute that script.
Continue with your own work
User guide follows a script or notebook from inspection to optional execution, comparison and sharing. Developer guide shows how to integrate the public compiler in your tools. Operations guide follows generated services through environments, verification and delivery. Use Examples for complete files and expected results, and Reference for exact options and contracts.
What the two MCP workflows do
The optional analysis MCP plugin lets a client inspect a repository with Apizr. A generated MCP server executes the Python functions you selected. You can call generated tools with the Python SDK without an AI account. The canonical concepts explain the shared terms.
Releases, Contributing, License and legacy compatibility remain available under Reference. The logo always returns to the home page.