Deploy an equation as an API in one command.
The confidence layer for AI-generated formulas.
pip install formulize
# Fit an equation to your data
import formulize
model = formulize.fit(df, target="y")
# See the equation + how trustworthy it is
print(model.equation) # y = 3.14 * sin(x1) + 0.5 * x2^2
print(model.confidence) # structure recurred in 847/1000 bootstraps
# Deploy as a live API endpoint (one command)
model.deploy() # → https://api.formulize.dev/e/abc1233-8× faster than PySR, 40% higher accuracy on Feynman equations.
| Engine | Time to fit (Nguyen suite) | Feynman equations recovered | Confidence bounds |
|---|---|---|---|
| formulize | 8.4 s | 88 / 100 | ✓ bootstrap-structural |
| PySR (Julia backend) | 34 s | 63 / 100 | ✗ (coefficient CI only) |
| Operon (C++ backend) | 67 s | 61 / 100 | ✗ |
| gplearn | 142 s | 41 / 100 | ✗ |
Wall-clock, single RTX 4090, Python 3.11 + Julia 1.10, batch size 100 problems. Reproduce in 60 seconds →
Three things nobody else ships
1. GPU-native from the ground up
The search runs on your GPU, not one CPU core. Handles 100K rows × 200 features without breaking a sweat.
2. LLM-typed-program proposer
Before the search starts, an LLM proposes candidate equations as typed programs — with dimensional units, monotonicity, and symmetry constraints. A cheap solver rejects the physically-impossible 99% before you spend a single CPU cycle fitting them.
3. Bootstrap-structural confidence
Every equation ships with a machine-readable certificate: "this structure recurred in 847 of 1000 bootstrap resamples." That's the difference between an alpha you can bet capital on and a data-mined artifact.
Pricing
Cancel anytime. All tiers include the full engine + confidence layer.
Team
- 10 seats
- Unlimited endpoints
- Unlimited fit-calls
- White-label PDF reports
- Priority Slack support
Enterprise
- Unlimited seats
- Custom risk factors
- Dedicated instance
- SLA + SOC 2
- Named analyst access
The core engine is MIT-licensed and free forever. Paid plans get the hosted `formulize deploy` platform, priority support, and enterprise features. Install it locally →
Want to be a design partner?
We're onboarding the first 20 users personally. Tell us what you're trying to fit and we'll set you up with a free 60-day pilot.