Tools

Tools#

EngiAI provides a comprehensive set of tools that agents use to complete tasks. All tools are LangChain @tool decorated functions that agents call automatically based on user requests.

ArXiv Tools#

Search and analyze academic papers.

search_arxiv#

Search for papers on arXiv.

from src.tools import search_arxiv

results = search_arxiv.invoke({"query": "topology optimization", "max_results": 5})

get_arxiv_paper#

Retrieve paper details by arXiv ID.

from src.tools import get_arxiv_paper

paper = get_arxiv_paper.invoke({"arxiv_id": "2301.12345"})

Additional tools (created via factory with MMORE client):

  • download_and_analyze_paper — Download PDF and index in MMORE

  • ask_about_papers — Query indexed papers via RAG

  • list_analyzed_papers — List all indexed papers

RAG Tools#

Retrieve information from your knowledge base using MMORE.

MMOREClient#

The MMORE client provides direct access to the document retrieval service:

from src.tools import MMOREClient

mmore = MMOREClient()
results = mmore.retrieve(query="What is topology optimization?", max_matches=5)

RAG tool functions (created via factory)#

from src.tools import create_rag_tools, MMOREClient

mmore = MMOREClient()
tools = create_rag_tools(mmore)
# Returns: [search_documents, add_document, add_url_to_knowledge_base,
#           list_documents, delete_document]

HPC Tools#

Interact with high-performance computing clusters via SSH/SLURM.

submit_slurm_job#

Submit a SLURM job script to the cluster.

from src.tools.hpc import submit_slurm_job

result = submit_slurm_job.invoke(
    {
        "slurm_file": "#!/bin/bash\n#SBATCH --job-name=test\npython train.py",
        "host_alias": "euler",
    }
)

get_slurm_job_status#

Check the status of a submitted job.

from src.tools.hpc import get_slurm_job_status

status = get_slurm_job_status.invoke({"job_id": "12345"})

Other HPC tools#

  • test_hpc_connection — Verify SSH connectivity

  • cancel_slurm_job — Cancel a running job

  • download_job_outputs — Retrieve job output files

Job monitoring tools (src/tools/job_monitor.py)#

Polling-based monitors that sit on top of get_slurm_job_status:

  • monitor_job_until_complete — Poll a job until it finishes (or max_checks is reached) and optionally auto-download outputs

  • check_job_status_change — Compare a job’s current state against a cached status and return whether it changed

  • get_active_jobs_summary — Summarize all jobs currently tracked in the in-memory status cache

Engineering Tools#

Optimization and design tools using EngiBench.

Supported problem types:

  • beams2d: 2D structural topology optimization

  • photonics2d: 2D photonic device topology optimization

(thermoelastic2d is still registered as a benchmark problem — see benchmarks/shared/problem_registry.py — but is not supported by these tools, so an evaluation against it fails at create_problem.)

optimize_design#

Run topology optimization on an EngiBench problem.

from src.tools import optimize_design

result = optimize_design.invoke(
    {
        "problem_type": "beams2d",
        "volfrac": 0.3,
        "forcedist": 0.5,
    }
)

All EngiBench tools#

  • create_problem — Initialize a problem instance

  • simulate_design — Evaluate a design’s objectives

  • optimize_design — Run topology optimization

  • render_design — Visualize the design as an image

  • get_problem_details — Get detailed problem description

  • get_dataset_info — Get dataset information

EngiOpt tools#

Tools for working with trained generative models:

  • download_wandb_model — Download a model from W&B

  • load_wandb_model — Load a downloaded model

  • sample_designs_from_model — Generate designs from a trained model

  • generate_training_command — Generate SLURM training scripts

  • evaluate_model — Evaluate a trained model against baselines

Algorithm registry (src/tools/algorithms.py)#

Single source of truth for supported generative algorithm IDs. Add an entry to SUPPORTED_ALGORITHMS (and to the AlgorithmId Literal) to expose a new algorithm to every EngiOpt tool automatically.

  • AlgorithmId — Literal type of valid algorithm IDs (currently cgan_cnn_2d, diffusion_2d_cond)

  • SUPPORTED_ALGORITHMS — Dict of algorithm metadata (class, dimensions, conditional, model type, model file names)

  • ALGORITHM_IDS — List derived from AlgorithmId, used by the EngiOpt tools for validation

Export Tools#

convert_design_to_stl#

Convert a 2D design to STL format for 3D printing. Uses xeltofab’s marching-cubes extruder, which produces watertight-by-construction meshes with smoother surfaces than the previous voxel-block backend.

from src.tools import convert_design_to_stl

result = convert_design_to_stl.invoke(
    {
        "npy_file_path": "outputs/design.npy",
        "scale_xy": 1.0,
        "scale_z": 10.0,
        "mirror_y": False,
        "threshold": 0.5,
        # Optional xeltofab pre-processing:
        "smooth_sigma": 0.0,  # Gaussian smoothing sigma (0 = off)
        "fill_holes": False,  # close single-pixel pinholes
        "min_component_area": 0,  # drop disconnected blobs smaller than this many pixels
    }
)

CLI Tools#

  • execute_cli_command — Run shell commands locally

  • open_gui_application — Launch GUI applications (PrusaSlicer, VS Code, etc.)

Search Tools#

  • create_search_tool() — Factory that returns a provider-neutral web_search tool (DuckDuckGo by default; Tavily optional)

Human Input Tools#

Tools that let an agent pause and request clarification from the user.

ask_human_for_clarification#

Used by the engineering agent when a user’s request does not specify required numerical design parameters (e.g., volume fraction, force distance, filter radius). The agent calls this tool instead of guessing defaults; the UI surfaces the question and waits for a reply before any design tool runs.

from src.tools.human_input import ask_human_for_clarification

ask_human_for_clarification.invoke(
    {
        "clarification_request": (
            "What volume fraction should I use for this beam design? "
            "Please specify a value between 0.1 and 0.9."
        ),
    }
)

Next Steps#