Introduction

When developing agent applications based on DeepSeek Harness (DSH), directly invoking simulation toolchains usually requires writing scripts to handle geometry modeling, meshing, solver configuration, and post-processing. dsh-cae-plugin (codename Mochi) provides an automated solution that can directly translate user-entered natural language instructions into a complete workflow covering CAD modeling, meshing, solving, and result verification.

What It Is

This is a DSH workflow plugin maintained by DaiYuhangSustc. It integrates tools such as build123d, Gmsh, CalculiX, and OpenFOAM, covering the entire process from CAD geometry processing to mesh independence verification. Its core positioning is a “natural-language-driven CAE pipeline,” supporting structural finite element analysis (FEA) and fluid dynamics (CFD) simulation.

Core Features

The plugin provides five core tool modules, corresponding to different stages of the simulation process:

  1. cae_geometry: Handles geometry processing. Supports building geometry with build123d scripts or importing external STEP files, including geometry validation and repair.
  2. cae_mesh: Handles mesh generation. Supports solid tetrahedral meshing (tet4/tet10) and fluid duct meshing.
  3. cae_solve: Handles solver computation. Supports CalculiX linear static solving, as well as OpenFOAM steady-state and transient incompressible CFD solving.
  4. cae_post: Handles post-processing of results. Uses PyVista for result extraction, plotting, and visualization.
  5. cae_verify: Handles verification checks. Provides automated mesh independence verification, including Richardson and GCI calculations.

Installation and Activation

Installing with the Docker image is recommended, without requiring manual environment configuration.

Docker installation:
Specify the image in the DSH configuration:

python: docker://ghcr.io/daiyuhangsustc/dsh-cae:latest

Local environment installation:
If not using Docker, you need to configure the Python environment yourself and install the following dependencies:
* pip install build123d gmsh pyvista ccx2paraview
* Install the solver: on Debian/Ubuntu use sudo apt install calculix-ccx, or install it via conda in other environments.
* Install OpenFOAM (Foundation v11–13 or the ESI version), and ensure its environment variables are detectable or specified through openfoamBashrc.
* Finally, load the plugin with a command (using the web profile as an example):

dsh plugin --profile web add /path/to/dsh-cae

Typical Usage

The plugin supports two operating modes: the Web interface and the command line.

Web mode:
Start the service:

dsh web

The service will start at http://127.0.0.1:3080. Paste a natural language instruction to view tool calls and result feedback.

Command-line mode:
Execute tasks in headless mode:

dsh --profile headless "20×20 mm square duct, 1 m long, water at 0.02 m/s inlet — steady laminar solve and a pressure contour"

The official examples provide two reference files:
* examples/cantilever.md: Static analysis of a cantilever beam.
* examples/duct-flow.md: Steady-state duct flow solving.

Applicable Scenarios and Precautions

This plugin is suitable for developers who need to quickly validate simulation concepts or perform repetitive simulation workflows using natural language. Note that the plugin depends on a specific computing environment (Docker, or locally installed build123d, gmsh, pyvista, calculix, and openfoam). In addition, if DSH Surface was started before the plugin was added, the plugin will not be detected and Surface must be restarted.

Conclusion

By wrapping complex CAE workflows behind natural language interactions through a modular toolset, this plugin lowers the barrier to simulation development. For more details, see the GitHub repository or the community catalog.