A free and open source Python library for computer-assisted mass appraisal (CAMA) — built for county assessors, appraisal districts, and property tax researchers who need valuations they can defend and reproduce.
pip install openavmkit
Download spatial variables from public sources and automatically tag parcels with them.
Get street adjacency from OpenStreetMap, building footprints from Overture, statistics from the US Census, topography from USGS, and more.
Automatically calculate distances between parcels and landmarks you care about—CBDs, universities, golf courses, and beyond.
Build automated valuation models using MRA, GWR, LightGBM, XGBoost, CatBoost, layered comparable sales, and more—all through one unified interface.
Ensemble multiple model results together for maximum performance.
See what drove every prediction: SHAP contributions and per-feature parameters are written out for each model, ensembles included, so no value is a black box.
Generate IAAO-compatible ratio studies and other statistical reports in PDF and HTML formats.
Measure horizontal and vertical equity, including the Vertical Equity Index, and report COD, PRD, and PRB alongside them.
Break down reporting by location, property type, price decile, or any other criteria you define.
Identify variables and characteristics that have the greatest impact on price.
Detect anomalous sales that undermine your predictive models—price outliers within clusters of similar properties, and suspicious price clusters.
Flag multi-parcel deals and misclassified vacant sales from deed IDs, sale dates, and prices.
Exclude invalid sales with legible and defensible rules you write down once, rather than one-off judgment calls.
Set up a clean pipeline that loads, transforms, and merges data without relying on one-off spreadsheet operations.
Define a reproducible set of steps once, then re-run the whole workflow end-to-end as often as you like.
The parts of the job that generic modeling tools skip, and that you get asked about at hearings.
Catch non-arms-length sales before they poison the model: spot potential multi-parcel deals by deed ID or same-date-and-price, parcels "sold vacant" with a building older than the sale, and price outliers within homogenous clusters.
Are similar properties in similar locations valued alike? Cluster comparable parcels and measure the horizontal dispersion of valuations within each cluster (CHD).
Are low-value and high-value properties valued with the same accuracy? PRD and PRB with bootstrap confidence intervals, per-quantile median ratios, and the new VEI stat.
Combine models into a prediction that beats any one of them alone, and get reassembled per-feature value contributions for the ensemble itself.
Build a market index from your own sales and adjust prices to your valuation date, or load in an index you already trust.
Per-feature value contributions for every model, which you can use as adjustments in a comp grid or to answer "why is my value this number?"
Your models, settings, and data stay yours. When a contract ends or staff turns over, the pipeline doesn't walk out the door with them.
Your state oversight body, an outside reviewer, or opposing counsel can re-run the whole roll from your settings file and inputs and get the same answer.
Full modeling, analysis, and GIS capability with no per-seat, per-parcel, or annual fees. Free under the AGPLv3, with a commercial license available if you need one.
Download the latest release and join a growing user community.
Get Started