Tools Reference

PumasAide provides three tools that AI assistants use to perform pharmacometric analyses. The AI writes Julia code directly against Pumas, NCA, AlgebraOfGraphics, and other Julia packages—guided by built-in documentation and workflow guides. You describe what you want in natural language, and the AI looks up the relevant APIs, writes code, and executes it on your behalf.

pumas_knowledge

Searches and reads built-in documentation covering pharmacometric concepts, API references, and analysis workflows. This is how the AI learns what code to write for your request.

The documentation includes:

  • API references for Pumas, NCA, DataFrames, AlgebraOfGraphics, SummaryTables, and Bioequivalence
  • Workflow guides for NCA, population PK, simulation, report generation, and more
  • Domain knowledge on PK parameters, model diagnostics, study designs, and regulatory considerations

When you ask for an analysis, the AI reads the relevant workflow guide and API docs before writing any code. You can also ask the AI to look things up directly: "Search the docs for bioequivalence acceptance criteria" or "What parameters does NCA.run_nca return?"

include_julia_file

Executes Julia code from files in your workspace. This is the primary execution tool—every analysis step produces a script in the programs/ directory that gets executed through this tool.

Supports .jl files and .qmd (Quarto) files with {julia} code blocks. Also supports executing specific line ranges within a file, useful for re-running part of an analysis after editing.

eval_julia_code

Runs Julia code for quick inspection and debugging. Use cases include checking variable types, previewing data, listing workspace objects, or examining intermediate values. The code isn't saved to a file—it's purely for exploration during your session.

Generated Code

Every analysis step produces an executable Julia script in the programs/ directory, organized by analysis phase:

programs/
├── .scratch/          # Development/debugging (gitignored)
├── 01-data/           # Data loading and preparation
├── 02-exploration/    # EDA, summary statistics, initial plots
├── 03-population/     # Population construction (Pumas or NCA)
├── 04-models/         # Model definitions
├── 05-fitting/        # Model fitting and estimation
├── 06-diagnostics/    # GOF plots, VPC, residuals
├── 07-results/        # Final tables, figures, reports
└── 08-sensitivity/    # Sensitivity analyses, dose linearity, BE

These scripts are fully reproducible—you can run them outside PumasAide, version control them with git, and review exactly what the AI did. The generated code uses standard Julia packages (Pumas, NCA, Bioequivalence, AlgebraOfGraphics, SummaryTables), so everything you learn transfers directly to manual workflows.