Real SCE-UA calibration
aiswmm calibrate now drives the spotpy engine, streams progress, guards unit mismatches, and writes auditable trial artifacts.
Latest v0.7.6 · real calibration + self-describing sessions ↗
Open source · verification first
Stormwater modelling with a chain of evidence.
Agentic SWMM turns a natural-language request into deterministic EPA SWMM execution, QA, provenance, calibration, and modelling memory—without hiding the files that support the result.
- v0.7.6 stable
- MIT licensed
- EPA SWMM 5.2.4
- PyPI + Docker
56typed tools
19reusable skills
3,060tests passing in v0.7.6
8documented evidence paths
Current release · 10 July 2026
v0.7.6 moves the runtime from demonstration to disciplined operation.
The latest release makes calibration real, every session self-describing, and every new run follow one canonical artifact layout.
Read the changelog →Self-describing sessions
Goals, provider and model, prompt and tool hashes, skill hashes, permissions, environment, and lifecycle travel with every session.
One canonical run layout
Builder, runner, QA, plot, audit, upstream, and review artifacts now land in stable numbered stages without breaking historical runs.
The method
Ask naturally. Execute deterministically. Keep the record.
This is not a chat wrapper around SWMM. The agent coordinates; the solver, checks, and artifacts remain inspectable.
Source or build the model
Bring an existing .inp, build from prepared GIS, or fetch Canadian storm-network data through SWMMCanada, Zhonghao Zhang's companion open-source project.
Run, calibrate, and verify
EPA SWMM runs stay CLI-runnable. QA gates, real SCE-UA calibration, unit guards, and direct solver comparisons keep automation grounded.
Audit, remember, refine
Provenance and experiment notes feed curated and raw memory. Skill changes remain proposals until a human reviews and benchmarks them.
Owned upstream project
SWMMCanada connects the workflow to Canadian storm networks.
Agentic SWMM accepts existing models and prepared GIS inputs. For Canadian network sourcing, it integrates with SWMMCanada and carries the retrieved archive and service identifiers into the run record.
Evidence, not theatre
Every visual has a stated boundary.
Benchmarks, research previews, and audit examples are presented with what they prove—and what they do not.
Project overview
See the workflow, then inspect the implementation.
The short introduction explains the modelling loop. The repository contains the runtime, 19 skills, MCP interfaces, tests, benchmarks, and audit contracts.
Published research · AI for Engineering · 2026
Auditable and reproducible stormwater modelling with Agent Skills and MCP.
Zhang, Z. & Valeo, C. · Volume 1, Issue 1, Article 5
Read the paper ↗Go deeper
Follow the question you have.
01
Why this exists
The modelling problem and the human-control boundary.
Vision → 02How it works
Runtime, skills, evidence contracts, memory, and calibration.
Features → 03What is supported
Benchmarks, research previews, and explicit evidence limits.
Validation → 04How to run it
Installer, PyPI, Docker, and portable skill setup.
Install →