ICortex is a Jupyter kernel that provides a soft-programming environment, allowing anyone to create programs with less formal and precise language. Prompts written in natural language are used to generate Python code at runtime.
ICortex overloads the regular IPython kernel with magic commands that provide code-generation capabilities and fine-grained control over generation context. As such, ICortex can run an existing IPython notebook without any compatibility issues. Cells that contain regular Python are executed in the current namespace as usual.
Code generation and execution history is saved in hidden variables and is used to construct the context for each new code generation. In other words, API calls to generate code in a cell contain information about previously ran prompts, executed cells, their outputs and other metadata related to the notebook.
%prompt magic is used to describe what the generated code should perform.
Running the command passes the prompt to
Each service defines how to parse the prompt and generate code
individually—each class that derives from ServiceBase
to parse the prompt and retrieve parameter values.
%prompt This is a prompt and will be used to generate code
The string that follows the prompt is parsed by argparse.ArgumentParser:
%prompt You can change how a prompt is run by providing flags -r --max-tokens 128
You can escape flags by surrounding the prompt text in single or double quotes:
%prompt "This flag -r is parsed verbatim"
%help prints a list of all the options you can generate code with your
ICortex provides the alias magic
%p, to let you write
%p This is the same as calling "%prompt ..."