jit-python

Note: This project is a proof of concept and is not intended for production use.

Dynamically generate and improve Python methods at runtime using an AI agent powered by Google ADK.

Inspiration

This project is inspired by luai.nvim by @tjdevries, which applies the same concept to Lua in Neovim. Watch the original video here.

How it works

JitPython intercepts calls to methods that don't exist yet and asks a Gemini-backed agent to write them. The generated code is compiled on the fly, bound to the instance, and called — all transparently. Describe what you want once, and the method is yours. Not happy with the result? Pass feedback to improve_method and the agent will revise it.

If the generated method requires a third-party package, the agent will emit code to install it via pip at runtime.

Security warning: JitPython executes AI-generated code and can install arbitrary pip packages at runtime. Run at your own risk.

Installation

pip install git+https://github.com/joshtyf/jit-python.git

Or clone and install locally:

git clone https://github.com/joshtyf/jit-python.git
cd jit-python
pip install -e .

Requirements

A GEMINI_MODEL environment variable can be set to control which Gemini model is used (defaults to gemini-3-flash-preview). A valid Google API key must be configured in your environment.

Usage

from jit_python import JitPython

jit = JitPython()

# --- Example 1: No external libraries required ---
# The agent generates pure Python code with no imports needed.
result = jit.fizzbuzz(
    15,
    __description="Return 'FizzBuzz' if divisible by 3 and 5, 'Fizz' if by 3, 'Buzz' if by 5, else the number as a string."
)
print(result)  # 'FizzBuzz'
print(jit.fizzbuzz(7))  # '7'

# --- Example 2: Builtin standard library required ---
# The agent imports from Python's stdlib (e.g. 're') inside the function body.
emails = jit.extract_emails(
    "Contact us at hello@example.com or support@foo.io",
    __description="Extract all email addresses from a string and return them as a list."
)
print(emails)  # ['hello@example.com', 'support@foo.io']

# Refine with feedback
jit.improve_method("extract_emails", "Also validate that each address has a valid TLD before returning it.")
print(jit.extract_emails("reach me at jane@bar.com"))  # ['jane@bar.com']

# --- Example 3: External library required ---
# WARNING: The agent will install the externals package (e.g. 'requests') via pip at runtime.
# Only run this if you trust the generated code and accept the security implications.
html = jit.fetch_webpage(
    "https://example.com",
    __description="Fetch the HTML content of a URL using the requests library and return it as a string."
)
print(html[:200])

# Inspect the generated source for any example
print(jit.get_method_code("fetch_webpage"))

API

Method	Description
JitPython()	Instantiates the agent. Emits a UserWarning about runtime code execution.
jit.<method_name>(*args, __description=..., **kwargs)	Calls a dynamically generated method. On first call, __description is required. Subsequent calls use the cached compiled method.
jit.improve_method(method_name, feedback)	Re-generates a method using the original session context plus the provided feedback.
jit.get_method_code(method_name)	Returns the raw source code of a previously generated method.