openpaso 1.3.0 — openpaso doctor and openpaso install

openpaso doctor shows what this install can do -- Python, openPASO, every solver, the mesh generator -- without a key or network. openpaso install <solver> checks first and installs only what is missing, the same way the server's setup_backend tool would; for the codes without a package (4C, deal.II, FEBio, SPARTA) it says what to do by hand. A named Kratos interpreter now runs with its own environment. Otherwise identical to 1.2.1.

openpaso 1.2.1 — scikit-fem comes with the default install

pip install openpaso now includes scikit-fem, so a fresh install has a working solver out of the box. The README badge row links PyPI and the install page describes the pip route first. Otherwise identical to 1.2.0.

openpaso 1.2.0 — first release on PyPI

openPASO 1.2.0 — first release on PyPI

pip install "openpaso[skfem]"     # the server plus the easiest solver to start with
openpaso                          # starts the MCP server on stdio

Point your AI app at the openpaso command (Claude Code: claude mcp add openpaso -- openpaso;
Claude Desktop / Cursor: "command": "openpaso" in the MCP configuration) and it can prepare,
run, couple and verify simulations in the nine supported codes. The pip-installable solvers come
as extras ([skfem], [ngsolve], [kratos], [dune], [all-solvers]); 4C, deal.II, FEniCSx,
FEBio and SPARTA are found on the machine as before. The same install is importable as
openpaso for scripts.

The project now lives at https://github.com/open-PASO/openPASO (moved from Hereon-InstituteMS;
the old URL redirects). Documentation: https://open-paso.github.io/openPASO/

Since 1.1.0

• Every solver survey prints verdict lines: a reference the run never saw, the measured value, a
  tolerance, PASS or FAIL — and each verdict kind has been shown to fail on a planted wrong answer.
• Coverage is reported against each backend's own declared surface (every FE class, space,
  element, registry entry, command) in four states: not attemptable, failed, ran, verified.
• Corrections found by those verdicts: the scikit-fem survey's boundary treatment for
  Hermite-type elements, its mixed-element quadrature mismatch, the FEniCSx Hermite
  interpolation text, DUNE element counts after refinement, an advertised scikit-fem row that
  could not be served, and a Kratos interpreter that was auditioned rather than used.
• The verification gate no longer marks every 4C VTK run unverified; the 4C FSI deck no longer
  writes an unfillable pressure; the equation check judges cell midpoints only and carries the
  reaction term; the coupling audit computes it per side; a level that never coupled is not
  counted as converged; a verified coupling is judged before its record is shortened.
• An installed package writes its state (session journal, discovery cache, build logs,
  community submissions) to the user's state directory, never into site-packages.

OASiS v1.1.0 — rebrand release

First release under the OASiS name (Open Agentic Simulation System). Capabilities are unchanged from v1.0.0; this release tags the post-rebrand state and is the version cited in the IJNME paper submission.

DOIs

• v1.1.0 archive: 10.5281/zenodo.20543501
• Concept DOI (all versions): 10.5281/zenodo.20388034

What's new in v1.1.0

• Rebrand: Open FEM Agent → OASiS (full sweep across README, CITATION.cff, AGENTS.md, CLAUDE.md, mcp_config.json, pyproject.toml, all backend logger names, and developer-mode template paths).
• Logo: added logo/logo.png and logo/logo_w_text.png.
• Version metadata: bumped CITATION.cff and pyproject.toml to 1.1.0.

Capabilities (unchanged from v1.0.0)

• 7 working FEM backends: FEniCSx, deal.II, 4C Multiphysics, NGSolve, scikit-fem, Kratos Multiphysics, DUNE-fem.
• 13 consolidated MCP tools.
• 26 cross-backend collation topics, 45+ pitfalls.
• 24 E2E development-corpus stress tests passing.

Citation

@software{hermann2026oasis,
  author    = {Hermann, Alexander and Shojaei, Arman and Scheider, Ingo and Cyron, Christian},
  title     = {{OASiS}: an open-source multi-physics and multi-code framework for verified computer simulations},
  month     = jun,
  year      = 2026,
  publisher = {Zenodo},
  version   = {v1.1.0},
  doi       = {10.5281/zenodo.20543501},
  url       = {https://doi.org/10.5281/zenodo.20543501}
}

Note: the v1.0.0 DOI (10.5281/zenodo.20388035) was a pre-rebrand test record and has been retracted by the author.

v1.0.0

First public release of Open FEM Agent.

What's in this release

• MCP server connecting any LLM coding agent (Claude Code, Cursor, Windsurf, …) to seven independently developed finite-element solvers: FEniCSx, deal.II, 4C Multiphysics, NGSolve, scikit-fem, Kratos Multiphysics, DUNE-fem. 180 solver–physics configurations out of the box.
• 13 MCP tools for solver discovery, simulation preparation, mesh generation, run execution, multi-solver coupling (domain decomposition, TSI, preCICE), in-place solver development, visualisation, and session insights.
• Curated pitfall knowledge base of common solver mistakes surfaced during 24 stress tests.
• Mandatory critic-sub-agent workflow that vets every simulation before it runs.
• Held-out benchmark (HoE-v1): 17 tasks × 4 conditions × 3 seeds = 204 cells released with per-cell results, statistics, and reproducible figures.
• 24 development-corpus stress tests released for transparency.
• MIT licence, CITATION.cff for citation.

Reproducibility

• 209 tests passing.
• All 204 benchmark cells released. Pooled pass rates: BARE 78.4%, MCP −PitDB 86.3%, MCP −Critic 88.2%, MCP Full 94.1%.

How to cite

See CITATION.cff for the software metadata. A Zenodo DOI for this release will be added once minted via the GitHub–Zenodo integration.