tmdl-lens

Automated documentation generator for Power BI projects.

Reads TMDL files from a folder of .pbip reports and generates structured HTML, Markdown, plain-text or JSON documentation for each one - covering data sources, tables, measures, relationships, M parameters, security roles, and calculation groups.

Runs as a standalone Windows desktop app or as a headless command-line tool. Output can be HTML (opens in any browser), Markdown (README.md for GitHub) plain text (<report>.txt for terminals, tickets and AI agents) or typed JSON (<report>.json for tools and CI). A live file watcher regenerates documentation automatically on every save.

What it documents

See a real example output generated from the included sample report.

Each generated README includes:

• Contents - a linked table of contents of the sections below (Markdown and HTML)
• Lineage (optional, Markdown) - a Mermaid source-to-table flowchart, see Lineage diagram
• Data Sources - connector type and resolved source detail for every query. 30+ Power Query connector functions have dedicated labels (SQL, Dataflows, SharePoint, Excel, OData, Azure Blob/Table/Data Lake storage, Databricks, Snowflake, Salesforce, Oracle, PostgreSQL, SAP HANA, and more); any connector without a dedicated label is still auto-detected from the query and shown with its namespace and function for manual labelling.
• Table Details - table type, source detail, and column list
• Measures - full DAX (optional), display folder, and format string
• Functions - DAX user-defined functions with description, where they are used, and full DAX (optional)
• Relationships - cardinality, cross-filter direction, and active/inactive state
• M Parameters - current value and which source expressions reference each parameter
• Security Roles - static and dynamic RLS rules per table, flagged by type
• Calculation Groups - calculation items with DAX and ordinal order
• Model Statistics - table, column, measure, and relationship counts

Installation

Option A - Command line (PyPI)

Requires Python 3.12+.

pip install tmdl-lens
tmdl-lens --reports-folder C:\path\to\reports --format md

The base package installs the dependency-free tmdl-lens command and nothing else. For the desktop GUI, install the GUI extras and launch it as a module:

pip install "tmdl-lens[gui]"
python -m tmdl_lens.app

Option B - Standalone executable

Download tmdl-lens.v<version>.zip from the Releases page, extract it, and run tmdl-lens.exe from the extracted folder. No Python or pip required. Keep the extracted files together - the executable loads its dependencies from the _internal folder next to it.

A config.json file will be created alongside the executable on first run to store your settings.

Note: Windows may show a SmartScreen warning on first run. Click More info then Run anyway. This is expected for unsigned applications.

Option C - Run from source

Requires Python 3.12+.

git clone https://github.com/nox-magistralis/tmdl-lens.git
cd tmdl-lens
pip install -r requirements.txt
python main.py

Usage

1. Launch the app
2. Under Configure, set the Reports folder to the root of your .pbip projects
3. Set an Output folder if you want READMEs written somewhere other than next to each .pbip
4. Press Run Now to generate documentation for all reports
5. Enable Watch for TMDL changes to auto-regenerate on every save

Each report's semantic model folder is found in this order: the report listed in the .pbip file's artifacts, then that report's definition.pbir (datasetReference byPath), then a <pbip name>.SemanticModel folder next to the .pbip as a fallback when the first two are absent. A report whose definition.pbir uses byConnection is skipped, because it has no local model, and so is a byPath reference to a missing folder. A model folder without a definition/ TMDL folder (for example a model.bim project) is skipped with a message, and an existing output file is left untouched.

If no output folder is set, each README.md is written next to its .pbip file. If a separate output folder is set, each report gets its own named subfolder inside it - for example docs/SalesReport/README.md, docs/FinanceReport/README.md - so multiple reports never collide.

Optional documentation metadata (owner, team, refresh schedule) is set in the Configure tab and stored in config.json: next to the executable for the standalone app, in the repository root when run from source, and in %APPDATA%\tmdl-lens on Windows or ~/.config/tmdl-lens elsewhere for a pip install. A pip install that finds no file there copies an older config.json from its package folder once and leaves the old file in place. Values left blank are omitted from the generated output.

By default the app skips reports whose TMDL files and documentation settings have not changed since the last run - each report is hashed, and only reports that actually moved are regenerated. Turn this off with Skip reports with no TMDL changes in Configure.

On startup the app validates the config and logs a warning for an empty reports folder, a watch debounce outside 1-300, or an output format other than html, md, txt or json; path existence is checked when a run starts. A config.json that cannot be read is logged as an error and is not overwritten until you press Save; a value of the wrong type is logged with the default used instead.

Command line

The tmdl-lens command runs the same documentation pipeline without the GUI:

tmdl-lens --reports-folder C:\path\to\reports
tmdl-lens -r C:\path\to\reports -o docs --format html --overwrite

Run tmdl-lens --help for the full list of options. Exit codes: 0 when everything succeeds, 1 when a report fails or the reports folder does not exist, and 2 when --reports-folder is missing or the file passed to --config exists but cannot be parsed (a UTF-8 byte order mark is accepted). Settings can be read from a config.json with --config (CLI flags override file values) - this is also what makes --skip-unchanged persist its content hashes between runs. A --config path that does not exist is warned about and not created; relative paths inside the config resolve against the config file's folder. A value of the wrong type is warned about and replaced by its default for that run, and the file keeps your value. With --skip-unchanged, a report whose content changed is regenerated even when --overwrite is off.

Output formats

The JSON output is versioned by schema_version (currently 1.0) and is deterministic across machines: no timestamp or tool version, every list of named items sorted by name, LF line endings. Values are fixed codes rather than display text (kind, source.type, cardinality, cross_filter, security_filter, warning code), a source that could not be resolved is {"resolved": false, "reason": ...}, and source is null for a table without an M source. Every shared expression is listed under expressions, including those no table uses. With --no-include-dax, every expression is null; hidden-reference warnings are still reported.

Lineage diagram

--include-lineage (config key include_lineage, GUI option Include lineage diagram) adds a ### Lineage sub-section at the end of "1. Data Sources" in the Markdown output: a Mermaid flowchart from each external source, through the shared queries, to every loaded and not-loaded table. GitHub, Azure DevOps and the VS Code preview render it without any script, so the output stays offline. Helper functions, scalar helpers and parameters are left out. It is off by default, and with it off the output is unchanged. The diagram is Markdown-only: with --format html, txt or json the option has no effect and a warning is logged once per run (the JSON output already carries each source's chain and combine_sources). Node ids are derived from a SHA-256 of each node's identity, so the diagram is byte-identical across runs and machines, and adding a table does not rename the other nodes.

Requirements

The command-line core has no third-party dependencies (pure Python standard library). The desktop GUI adds two packages:

Python 3.12 or later required when running from source.

Compatibility & Public API

The items below are the public contract of tmdl-lens. They are frozen from version 1.0.0 - changing, renaming or removing any of them is a breaking change and requires a new major version, announced in the GitHub release notes.

CLI flags (tmdl-lens command):

--version, -r / --reports-folder, -o / --output-folder, --format (html, md, txt or json), --include-dax / --no-include-dax, --include-lineage / --no-include-lineage, --overwrite / --no-overwrite, --skip-unchanged / --no-skip-unchanged, --owner, --team, --refresh-schedule, --config, -q / --quiet.

Exit codes: 0 success; 1 a report failed or the reports folder does not exist; 2 --reports-folder is missing or the --config file exists but cannot be parsed.

config.json keys: reports_folder, output_folder, output_format, overwrite_readme, include_dax, include_lineage, skip_unchanged, watch_debounce, owner, team, refresh_schedule, file_hashes, features.watcher.

Not part of the contract: importing tmdl_lens.* as a Python library (the package is distributed for its CLI - the module layout may change) and the exact wording or layout of the generated HTML/Markdown (best effort).

Limitations

• Import-mode PBIP reports only. Live Connection, Direct Lake and other models without a definition/ TMDL folder are not documented; such a report is skipped with a message naming the reason.
• Connectors are detected from the query text. Known connectors get a dedicated label; anything else is shown with its Power Query namespace and function and is meant to be labelled manually.
• The parser targets the TMDL shapes seen in real PBIP projects, but the format keeps evolving - an unhandled object type is reported rather than crashing.

Development

pip install -r requirements-dev.txt
python -m pytest

requirements-dev.txt adds dev-only dependencies (pytest, pyinstaller) on top of the base requirements.txt. The test suite runs against the bundled sample report and covers the TMDL parser, source resolution, config validation, and README/HTML generation.

Project structure

tmdl-lens/
  main.py                   entry point (desktop GUI)
  pyproject.toml            package metadata and CLI entry point (PyPI)
  requirements.txt          GUI dependencies (source installs)
  requirements-dev.txt      dev-only dependencies (pytest, build, twine, pyinstaller)
  pytest.ini
  tmdl_lens/
    app.py                  CustomTkinter UI
    cli.py                  command-line interface
    pipeline.py             documentation generation pipeline
    tmdl_parser.py          TMDL file parser
    source_resolver.py      M expression source resolver
    readme_generator.py     README markdown generator
    watcher.py              file watcher (watchdog)
    config.py               app settings + documentation metadata (config.json)
    version.py              single source of the app version
  docs/                     TMDL syntax and parser internals reference
  sample/                   sample .pbip project for testing
  tests/                    pytest suite

Author

Marcin Mozol - github.com/nox-magistralis

License

MIT - see LICENSE