# install.packages("remotes")
remotes::install_github("CathalByrneGit/tardoc")
Nothing beyond targets and a handful of pure-R packages is needed for the
documentation itself. The analytics tiers are opt-in and their packages live in
Suggests.
From the root of a targets project:
tardoc::document_targets(pkg_name = "My pipeline")
tardoc::view_tardoc()
That writes the markdown, the self-contained viewer and the WASM
analytics page, then opens the viewer as file://. No server, no build step.
Useful arguments:
| Argument | Default | Description |
|---|---|---|
project_path |
"." |
Targets project root |
site_dir |
"tardoc" |
Output subfolder |
pkg_name |
"targets docs" |
Title in viewer headers |
pkg_desc |
"" |
Description for llms.txt |
repo_url |
NULL |
Repo base URL for source links, e.g. "https://github.com/user/repo/blob/main/" |
group_by |
"auto" |
Grouping signal for the graph — see Grouping the graph |
check_outdated |
TRUE |
Run tar_outdated() for freshness status |
llm |
FALSE |
Auto-generate missing descriptions and function explanations |
The full argument list, for every exported function, is on the manual page.
my_project/
├── llms.txt
└── tardoc/
├── viewer.html self-contained, opens as file://
├── wasm_analytics.html self-contained, opens as file://
├── analytics.html tier 3 shell, needs view_tardoc_db()
├── search_index.json
├── tardoc_analytics.json
├── history.json one snapshot per build, for the run diff
├── targets/
│ └── clean_data.md one .md per target
├── functions/
│ └── clean_raw.md one .md per function
└── notes/
├── index.md yours — the home-page overview
├── targets/clean_data.md yours — never overwritten
└── functions/clean_raw.md
tardoc.duckdb is not in that list. document_targets() does not build it —
view_tardoc_db() does, on first call.
Two conventions protect your own writing:
tardoc/notes/ on the first run and are
never touched again. Edit them freely; they are embedded into the viewer on
the next run.<!-- tardoc:generated -->
markers in the .md files. Anything you write outside them survives a re-run.llms.txt is written at the project root following
llmstxt.org — paste it into any LLM conversation for
pipeline context with no indexing step.
No. tar_manifest() and tar_network() only require _targets.R, so the full
documentation can be generated from a pipeline that has never been run.
If a store is present, build status and timestamps appear on target pages.
Without one every target reads Not built, which is accurate rather than a gap.
tar_outdated() re-hashes files and dependencies, so on a large pipeline it
costs a few seconds. Pass document_targets(check_outdated = FALSE) to skip it;
status then falls back to reporting errors only.
A dynamically branched target is a special case worth knowing about: its own
metadata row carries no build time — only its branches do — and its errors live
on the branch rows too. tardoc reads the branches, so a branched target that has
run reports Up-to-date, and one with a failed branch reports Errored rather
than hiding it.
Always required (Imports): targets, dplyr, purrr, roxygen2, Rd2md,
jsonlite, rlang, withr.
Optional (Suggests):
| Package | When needed |
|---|---|
duckdb, DBI |
Server analytics and tardoc.duckdb generation |
callr |
Reading _targets.R in a subprocess; the background Quack and MCP servers |
httpuv |
The local HTML server behind the server analytics viewer |
ellmer |
llm = TRUE at documentation time, or the chat tab |
callr is a suggestion rather than an import on purpose: where no R subprocess
can be spawned — webR, for instance — tardoc reads the pipeline in the current
session instead. See the webR note.