Jenner

Jenner is a SAS-compatible alternative to SAS: it runs most SAS programs out of the box -- DATA steps, procedures, macros, and SQL -- without a SAS license. Built from the ground up in Rust, it adds capabilities SAS doesn't have: native CSV and Parquet I/O, direct database connectors, a Jupyter kernel, ODS-compatible graphics, and PROC AI for LLM-powered data processing.

Statistical and time-series procedures (ARIMA, REG, LOGISTIC, LIFETEST, and others) use Python (statsmodels) or R (survival, forecast) as their computational backend while accepting standard SAS syntax.

Jenner reads and writes SAS7BDAT datasets natively. Reading uses vendored ReadStat; writing uses a native Rust port of FredHutch/sas7bdat producing files that round-trip through pyreadstat, R's haven, and Jenner's own reader. Those three are independent readers; the round-trip is verified against them, not against a SAS installation. SAS7BDAT is a first-class storage engine, not an export-only format: a DATA step, PROC SQL, or a LIBNAME declared with the engine all write it directly, exactly as they write Parquet or CSV. Avro is Jenner's default rather than SAS7BDAT — see storage engines for what that changes and how to work end to end in SAS7BDAT when you need to.

SAS Transport (XPT) read/write is also supported for FDA submissions. Legacy subsystems such as SAS/SCL and SAS/AF are not implemented.

See your data

Jenner ships a family of visualizer procedures — interactive 3-D and 2-D landscapes driven straight from a PROC step. PROC SPLATVIZ turns a dataset into a cloud of density you fly through:

A splat visualization of customer-churn data — a gas-like cloud of density inside a 3-D box

Scatter and splat landscapes, statistics panels, tree and map landscapes, decision-table and evidence cakes, cluster panes and association-rule floors — every one of them a PROC, each with a JSON view spec and an API an AI agent can drive.

Explore Visual Data Mining

Numerical accuracy

Jenner passes all 58 datasets in the NIST Statistical Reference Datasets benchmark — the standard third-party test suite for verifying statistical-software accuracy. That includes the high-difficulty linear-regression cases (Filip, Wampler1–5, Longley), nonlinear regression edge cases (Bennett5, ENSO, MGH09), and the f64-pathological SmLs07 / SmLs08 / SmLs09 ANOVA cluster — the canonical reproducer for catastrophic floating-point cancellation that bounds every double-precision statistical package at LRE ≈ 4 to 5. Jenner reaches LRE 15 on those three datasets (the f64 mantissa exhaustion point) through automatic precision recovery that engages without configuration when the data needs it.

The full per-dataset report, with Jenner's input scripts, computed outputs, and matched-against-NIST analysis for every test, lives at NIST StRD Accuracy. On this benchmark family Jenner's results sit above the historical hierarchy of every major statistical package — see the comparison section for citations against Excel, SPSS, Gretl, Stata, SAS, and R.

Quick Example

Pull US GDP from the Federal Reserve, population from the World Bank, merge them, and forecast GDP per capita with ARIMA:

/* Connect to FRED and World Bank */
libname fred sasefred
    idlist="GDP" startdt="2000-01-01" enddt="2024-12-31";

libname wb sasewbgo
    idlist='SP.POP.TOTL'
    countrylist='USA' startyr=2000 endyr=2024;

/* Pull quarterly GDP and tag each quarter with its year */
data gdp;
    set fred.gdp;
    year = year(input(date, yymmdd10.));
    gdp = value;
    format gdp comma12.;
    keep year gdp date;
run;

/* Pull population */
data pop;
    set wb.sp_pop_totl;
    population = value;
    keep year population;
run;

libname fred clear;
libname wb clear;

/* Merge on year, compute per-capita GDP */
proc sort data=gdp; by year; run;
proc sort data=pop; by year; run;

data combined;
    merge gdp pop;
    by year;
    gdp_per_capita = gdp / population;
run;

/* Forecast 4 years with ODS chart output */
ods graphics on;

proc arima data=combined;
    identify var=gdp_per_capita(1) nlag=8;
    estimate p=1;
    forecast lead=4 out=forecast;
run;

The ods graphics on statement produces a forecast plot with observed values, fitted model, and 95% confidence intervals:

ARIMA forecast of GDP per capita

Getting Started

The fastest way to try Jenner is to sign up for Jenner Workspace and start writing programs in your browser. You can also download Jenner to run on your own machine — Windows, macOS, or Linux, each as a native binary (requires a license agreement — contact [email protected]).

Jenner Workspace

Sign up and start writing programs in your browser -- no installation required
Installation

Run Jenner in the cloud, or install the native binary on Windows, macOS, or Linux
First Program

Write and run your first Jenner program from the command line
Language Reference

Complete reference for DATA steps, procedures, functions, and macros
Data Access

Read and write CSV, Parquet, Avro and SAS7BDAT, and connect to PostgreSQL, Oracle, Snowflake and more
File Formats

How Jenner detects a format, and which engines it reads and writes

Why Jenner?

Capability	Jenner	SAS
Runs most SAS programs	Yes	Yes
Native CSV and Parquet I/O	Yes	No
Read SAS7BDAT datasets	Yes	Yes
Write SAS7BDAT datasets	Yes, natively	Yes
Mix SAS7BDAT, Parquet, Avro and CSV in one program	Yes	No
SAS Transport (XPT) read/write	Yes	Yes
Native binaries for Windows, macOS and Linux	Yes	No native macOS
No per-seat license required	Yes	No
Standalone binary	Yes	No
Native Jupyter kernel	Yes	No
Built-in database connectors	Yes	Add-on
ODS-style graphics	Yes	Yes
LLM integration (PROC AI)	Yes	No
Passes full NIST StRD benchmark (58 / 58)	Yes	Strong, with caveats on hardest f64 cases
Auto-recovers precision on stiff data	Yes	No
SAS/SCL, SAS/AF	No	Yes

See the full SAS alternative comparison for licensing cost, platform support, and syntax-compatibility details, the Altair SLC / WPS comparison if you run SAS-language code on SLC, or what's different from SAS for a syntax-level breakdown.

Need Help?

Documentation -- you are in the right place
Support -- [email protected]