Open source · Verification-first

Stormwater modelling you can reproduce, audit, remember, and trust

Agentic SWMM Workflow wraps EPA SWMM in the aiswmm runtime — natural-language orchestration, deterministic runs, QA checks, provenance, and modeling memory, with you in control.

Get started Read the paper
  • MIT licensed
  • EPA SWMM 5.2
  • aiswmm on PyPI
  • Stable v0.7.4
aiswmm
$ pip install aiswmm
$ aiswmm
✓ runtime ready · EPA SWMM 5.2
› "Run the Tecnopolo model, check peak flows"
  · deterministic SWMM run
  · QA + provenance recorded
  · audit note written to memory
How it works

Execute, remember, refine

01

Execute

Run SWMM deterministically. Docker and local scripts keep execution repeatable, emitting traceable artifacts at every stage.

02

Remember

Provenance and QA summaries become Obsidian-compatible modeling memory — inspectable, reusable, honest about failures.

03

Refine

Audited runs surface patterns and propose skill refinements — accepted only after human review and benchmark checks.

The ecosystem

Part of a larger agentic water-modeling stack

Agentic SWMM doesn't stand alone. The same natural-language, verification-first approach runs from building the model to executing it — and reaches across SWMM, MIKE+, and deep-learning hydrology.

Upstream · builds your model

SWMMCanada

Draw a boundary anywhere in Canada and get a complete, runnable EPA SWMM model. Built from Canadian open data — real municipal storm networks for eight cities, or a synthesized network anywhere else. It's the front door that hands a ready .inp to Agentic SWMM.

  • Map boundary → .inp
  • GeoPackage + netCDF
  • MIKE+ import package
  • Validation reports
Open SWMMCanada on GitHub →
MIKE+ engine

Agentic MIKE+

A headless, natural-language workflow for MIKE+. An MCP server that lets Claude, Codex, or Hermes inspect, edit, run, and visualize models — no GUI required.

Open on GitHub → Deep-learning hydrology

Agentic Hydrology Platform

Ask in plain language, get the right simulation. An LSTM streamflow engine and hub that coordinates SWMM, MIKE+, and ML behind one conversational interface.

Open on GitHub →
Project overview

See the Agentic SWMM workflow in video

The workflow

One trail from request to verified result

Agentic SWMM modeling memory and controlled skill evolution loop
5.2EPA SWMM engine
MITOpen-source license
PyPIaiswmm runtime
7Validation benchmarks

Explore the project

Vision →

Why a modeling workflow should remember, audit, and stay reviewable.

Features & Workflow →

The modular skill layer, verification-first provenance, and what a run produces.

Validation & Audit →

Runnable benchmarks, evidence boundaries, and the experiment audit framework.

Download & Install →

One-line installers, the pinned Docker image, and the aiswmm Python package.