Auto-generate documentation for any targets pipeline. Point tardoc at your project and get structured markdown, a browsable HTML viewer, and — optionally — a full analytics stack with SQL queries, semantic search and an LLM chat interface.
# install.packages("remotes")
remotes::install_github("CathalByrneGit/tardoc")
tardoc::document_targets(pkg_name = "My pipeline")
tardoc::view_tardoc()

Nothing in that screenshot is hand-written: every page, badge and graph comes
from _targets.R and the roxygen comments in R/. It is the example pipeline in
inst/examples/station-monitoring, published
on every push — open the
live demo.
document_targets() writes one .md per target and per function, plus two
self-contained HTML files that open as file:// — no server, no R on the
reader’s side:
Up-to-date, Outdated,
Errored or Not built, so a target invalidated by an upstream change is
visible before you run anything.llms.txt at the project root, for pasting a pipeline into an LLM
conversation.Three optional layers go further: a live DuckDB server with semantic search, an LLM chat tab, and the pipeline database as an MCP server for Claude Desktop and Claude Code.
A store does not have to exist. tar_manifest() and tar_network() read
_targets.R alone, so a pipeline that has never been run documents fully — with
every target reading Not built, which is accurate rather than a gap.
Full documentation is at cathalbyrnegit.github.io/tardoc:
clean+, +report, ingest_*MIT. See LICENSE.
To cite the package tardoc in publications, please use:
Byrne C (2026). tardoc: Generate Documentation and Analytics for targets Pipelines. R package version 0.6.0, https://cathalbyrnegit.github.io/tardoc/.
@Manual{,
title = {tardoc: Generate Documentation and Analytics for targets Pipelines},
author = {Cathal Byrne},
year = {2026},
note = {R package version 0.6.0},
url = {https://cathalbyrnegit.github.io/tardoc/},
}