PyAutoFit: Scientific Inference with Natural Language
AI Assistant | Documentation | Installation Guide | Introduction on Colab | HowToFit
PyAutoFit is designed so scientists can bring their models, data and likelihood code, then fit models, explore results and develop analyses using natural language with an AI coding agent.
Getting Started
PyAutoFit and the autofit_assistant allow one to perform scientific inference using purely natural language.
Simply open your AI coding agent (codex or Claude Code are recommended) and input the following prompt:
Example Natural Language Prompt for Claude Code, Codex or other AI coding agent
I want to perform scientific inference with PyAutoFit (https://github.com/PyAutoLabs/PyAutoFit) and the
autofit_assistant (https://github.com/PyAutoLabs/autofit_assistant).
Begin the "start here" guide for a new user.
PyAutoFit is designed to be used entirely through natural language — the natural-language inference page walks through this, including its AI First Design.
Bring Your Own Likelihood (BYOL)
Already have a likelihood function for your science problem? Point the assistant at your existing code and it can set it up with PyAutoFit — defining the model, choosing priors with you, configuring a search and organising the results:
Example Natural Language Prompt for Claude Code, Codex or other AI coding agent
Set up PyAutoFit with my existing science project. An example likelihood function can be found at
[GitHub link or local directory].
First, give me an overview of my project and likelihood function. Compose an appropriate model, explain it to me,
and recommend a non-linear search (for example MCMC, nested sampling or maximum-likelihood estimation).
Do not begin inference until we have discussed the setup and I give you the go-ahead.
Once inference is running, explain how the results are written to disk and show me how to inspect and interpret
them with PyAutoFit.
The assistant answers this prompt with its BYOL mode — bring your own likelihood, the assistant brings the inference. It reads your code and restates what your likelihood scores, composes a model with priors chosen with you, wraps and validates your function unchanged, recommends a search, and runs nothing until you say go.
Your existing science code remains the source of the likelihood. With PyAutoFit built around it, you can perform inference through natural language while gaining access to features such as flexible priors and model composition, MCMC and nested sampling, automated result handling, model comparison and scalable workflows.
What is PyAutoFit?
PyAutoFit is a domain-agnostic Python package for scientific model fitting and Bayesian inference. It supports nested sampling, MCMC and optimisation, alongside advanced methods such as hierarchical models, search chaining and Bayesian model comparison.
autofit_assistant connects natural-language requests to runnable PyAutoFit workflows. Ask it to compose a model, discuss priors, run inference or compare competing explanations. You guide the science; it writes and runs Python scripts you can inspect, rerun and share.
Human Readable Documentation
- The PyAutoFit readthedocs, which includes an installation guide and an overview of PyAutoFit's core features.
- The introduction Jupyter Notebook on Colab, where you can try PyAutoFit in a web browser (without installation).
- The autofit_workspace GitHub repository, which includes example scripts demonstrating PyAutoFit's features.
- The standalone HowToFit repository, a series of Jupyter notebook lectures which give new users a step-by-step introduction to PyAutoFit.
HowToFit
For users less familiar with Bayesian inference and scientific analysis you may wish to read through the HowToFit lectures. These teach you the basic principles of Bayesian inference, with the content pitched at undergraduate level and above.
The lectures are available in the standalone HowToFit repository.
Use the Jupyter notebooks if you want to run the code (recommended), or read the available Markdown lectures directly on GitHub.
For help alongside the lectures, open the autofit_assistant repository in your AI coding agent, following its setup instructions, and paste:
Enter HowToFit mode.
I want to work through the HowToFit lectures. Show me where to find them
and how to use Jupyter Notebook or Markdown, then help me with questions
as I go.
The assistant will answer questions about concepts, equations, code and results as you study, and help with notebook errors. Share the lecture link and section or the cell you are working on; you choose when to move on.
Community & Support
Questions, help with your code or your analysis, and ideas: the PyAutoLabs Discussions. Bug reports with a reproducer (a snippet, the traceback, your versions): an issue on the library's tracker. The Slack is for collaborators, by invitation.
Collaborators receive the latest PyAutoFit updates in the Slack channel. Contact James Nightingale about collaborator access.
Release files for autofit 2026.9.27.2
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | |
|---|---|---|---|
| autofit-2026.9.27.2.tar.gz | 8.5 MB | Details |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Reset |
|---|---|---|---|---|
| autofit-2026.9.27.2-py3-none-any.whl | Python 3 | none | any | Details |
Total release size: 17.1 MB
Release files / autofit-2026.9.27.2.tar.gz
| Download URL | autofit-2026.9.27.2.tar.gz |
|---|---|
| Size | 8.5 MB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
8e5ac0ee87eb252389e13fb025373702fd56bf22e6a6f71342c33e75335f94e5
|
|
BLAKE2b-256 checksum How to use checksums |
a3754ca92c316fae1976ff450cca32d4958f37fb7643d7c5d61276bd5e00ac69
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.0.1 CPython/3.12.14
|
Release files / autofit-2026.9.27.2-py3-none-any.whl
| Download URL | autofit-2026.9.27.2-py3-none-any.whl |
|---|---|
| Size | 8.6 MB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
e3c1d14e197724cb75e04a54010480952c76296eb841067ad23c7ac36bc97a0d
|
|
BLAKE2b-256 checksum How to use checksums |
5f689e77a3b212c67f971462269278132565054611a314ed4caf4e33189e4f3a
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.0.1 CPython/3.12.14
|
