PyPI agent tools

open-fdd on PyPI is the Python toolbox an AI agent (or an engineer) runs next to the Open-FDD app. Everything is SI and read-only toward the building system. Package versions (4.x) are separate from the app (5.x).

pip install open-fdd                       # core: ECM / ROI calculators and workbooks
pip install "open-fdd[nf,anomaly,ml]"      # add extras as needed (table below)

Tools

Tool What it does Extra Example
Rules (open_fdd.rules) pandas twins of the SQL fault rules, SI, same gates as the app oracle from open_fdd.rules import run_all
NF client (open-fdd-nf) Read an NF box: model, roles, SI history, run the rules, summary report nf open-fdd-nf rules --days 7
SPARQL to DataFrame (open-fdd-nf sparql, open_fdd.nf.sparql_data) SPARQL on the site model, matched points’ history as SI DataFrames, then the rules nf (rdf on Lite) open-fdd-nf sparql "SELECT ..." --days 7 --rows
NF analytics (open-fdd-nf analyze) Weekly runtime, cooling OAT bins, economizer, fault timeline, run on the box nf open-fdd-nf analyze weekly-runtime --days 28
Analytics (open_fdd.analytics) Motor hours, OAT bins, VAV health, metering and degree days, M&V change point, weather and psychrometrics analytics from open_fdd.analytics import motor_run_hours_table
Anomaly screen (open-fdd-anomaly screen) Rolling z-score, MAD, STL, Isolation Forest on a device folder, plots and scoreboard anomaly open-fdd-anomaly screen ./AHU_1 --out ./out
SQL z-score (--engine sql, open_fdd.analytics.sql_anomaly) The z-score as portable SQL over fdd.point + fdd.input_sample; equals the pandas flags anomaly (duckdb/postgres for those engines) open-fdd-anomaly screen ./AHU_1 --out ./out --methods zscore --engine sql
RCx report (open-fdd-anomaly report) Single-system or building FDD / RCx Typst report, PDF with --compile anomaly open-fdd-anomaly report ./AHU_1 --out ./rpt --month 2026-06 --compile
Engineering Findings (open-fdd-report) Evidence-reviewed findings from a checklist, an NF box or a CSV: DOCX, XLSX, JSON reporting open-fdd-report --csv export.csv --out ./findings --docx --xlsx
ML detectors (open-fdd-ml detect) PCA / SPE with contributions, gray-box residuals, mode-gated one-class SVM / COPOD, trained on known-good rows ml (ml-copod) open-fdd-ml detect ./AHU_1 --method grey_box --train-until 2026-06-01 --out ./ml
LBNL benchmark (open-fdd-ml benchmark) Scores the rules and detectors on public LBNL AHU data: TPR, FPR, delay (results) ml open-fdd-ml download && open-fdd-ml benchmark --out bench.md
Ontology audit (open-fdd-ontology-audit) Requirements sheet vs pinned Brick / 223P / Haystack snapshots (evidence-backed statuses); model doctor: standard vs Open-FDD vs site terms, identity, history links, SHACL ontology (ontology-shacl) open-fdd-ontology-audit req.xlsx --graph building.ttl --shacl --out r
Energy calcs (open-fdd-ecm calc) Fan affinity, schedule reduction, OA sensible / total, economizer cap, CHW reset, kW/ton, boiler efficiency, condenser water core open-fdd-ecm calc fan_affinity --json '{"design_kw":20,"hours":4000,"baseline_speed_fraction":1,"proposed_speed_fraction":0.8}'
ROI (open_fdd.ecm_engineering.finance) Simple payback and NPV with discount and escalation core npv(50000, 12000, years=15, discount_rate=0.05)
ECM workbooks (open-fdd-ecm demo, stage2-workbook) Auditable Excel ECM workbooks with dual-rail inputs and assumptions core open-fdd-ecm demo --out Demo_ECMs.xlsx
EnergyPlus evidence (open-fdd-ecm agent) Import EnergyPlus evidence; types InputRail.ENERGYPLUS, SourceType.ENERGYPLUS_AUTOSIZED / _DERIVED, provenance, honesty status (fitted vs ballpark) core open-fdd-ecm agent import_energyplus_evidence evidence.json
M&V (ecm_engineering.changepoint, g14) IPMVP change-point baselines, ASHRAE Guideline 14 NMBE / CVRMSE core from open_fdd.ecm_engineering import g14
Parity (open_fdd.parity) Seeded SI dataset and the pandas == SQL comparison oracle python tests/parity/stress.py --engines duckdb
Version (open-fdd-version) Package and app versions as JSON core open-fdd-version --pretty

Agent skills for these tools: openfdd_agent_spec/skills/. More detail: python-tools.md. The 4.x ECM guide (engineering calcs, IPMVP, EnergyPlus purpose) was retired from the site in 5.0 and is kept in git history.