An interactive curve-fitting application. You load a data set, place curves on it, and the app fits them — one peak or a hundred, by hand or fully automatically.

Written in Free Pascal with Lazarus. It runs on Linux, Windows and macOS.

Automatic decomposition

One command, Fit → Automatically, turns a raw profile into a small set of curves. It subtracts the background and seeds a curve on every point of every peak. It then removes curves one at a time for as long as the fit stays within the accuracy you set, the maximum acceptable R-factor. What is left is the fewest curves that accuracy allows. You don't guess the number of curves or tune a penalty weight. On the two-peak sample that ships with Fit, 32 starting curves come down to 3 in well under a minute.

How many curves does the profile need? — the rule, and how it compares with AICc, BIC and the other published methods.

Download

Built by CI from the sources in this repository. On macOS, build it yourself: one script produces Fit.app and installs it.

Each link above serves the latest release · all releases

Fit is two programs: the desktop client, which has no fitting engine of its own, and the compute server it talks to over HTTP. Every installed Fit starts the server itself — the Windows shortcut, the Linux fit command and Fit.app all launch through a wrapper that starts a server on port 8787 when nothing answers there, and reuses one that does. The portable Linux archive is the exception: start fit_server before the client.

Building from source covers the script, the macOS bundle and a Lazarus IDE walkthrough with exact versions.

Watch it work

Where it stands

Using it

• The fitting workflow — fitting by hand, fitting again, and automatic decomposition
• Project files — save a session and carry on later
• Curve types
• User-defined curves — your own formula, no code
• Argument axes and units
• Loss functions — what “best fit” means here
• Compute backends — native, Python/lmfit, or a remote server
• Building from source

Extending it

Fit is a framework as much as an application. A new curve type, data loader, optimiser, objective, REST verb or whole analysis vertical is added by registration — a directory plus one entry on a project's unit search path. No framework file changes.

These pages are generated from the registries this build actually contains, every time the site is published:

• Architecture — the three processes, the client-to-server call chain, how a running fit reports progress, the view seam, all ten extension seams, and the anatomy of a module
• Extension points — one section per seam: what you write, where it goes in, and what is registered through it today
• Adding a curve type — the registration path, the curve classes, and curve types the user defines
• Adding a data loader

The prose that goes with them lives in the repository: writing a module · a complete working module in six files · AGENTS.md, the invariants, for AI agents working on the code.

Command line

The client and the server keep separate logs — fit_client.log and fit_server_log.txt — in $HOME/Fit/ on Linux and macOS, or C:\Users\<user>\AppData\Roaming\Fit\ on Windows. Each rotates at 32 MB, keeping one previous generation as <name>.1.

Test data

A sample data set ships inside every download, under Data/. Separately: test_data.zip.

Contributing

This repository is published as a snapshot: each publication replaces main with a single commit, so a pull request cannot be merged directly. Open an issue or attach a patch — see CONTRIBUTING.md.

GPLv3-or-later. Documentation CC BY 4.0.