Commands

Ten commands. Only two of them call an LLM.

Command	What it does	Calls an LLM
`init`	Write `.deepdoc.yaml`	No
`generate`	Scan, plan and write the docs	Yes
`update`	Refresh docs after code changes	Yes, for changed pages
`serve`	Preview locally	No
`deploy`	Build a static site	No
`status`	Show what exists and what is stale	No
`config`	Read or write config values	No
`performance`	Timing and token metrics	No
`clean`	Remove everything DeepDoc created	No
`benchmark`	Score planner quality	Depends

`init`

Creates .deepdoc.yaml in the current repository.

deepdoc init --provider anthropic

--provider is required. Everything else has a sensible default.

Flag	Description
`--provider`	`anthropic`, `openai`, `azure`, `ollama`, or any LiteLLM alias
`--model`	Defaults to a sensible model for the provider
`--name` / `--description`	Project name and description; default to the directory name
`--output-dir` / `--site-dir`	Default `deepdoc-docs` and `deepdoc-site`
`--with-chatbot`	Also scaffold the optional Q&A backend
`--llm-max-concurrency` / `--llm-rpm` / `--llm-tpm`	Rate limits
`--context-window-tokens`	Override the detected context window

For Google Gemini, run init with any provider and then set the model with deepdoc config set. See Providers.

`generate`

Scans your code, plans the structure, and writes Markdown. This is the step that costs LLM calls.

deepdoc generate

Refuses to overwrite existing DeepDoc-managed docs unless you say so:

deepdoc generate --force   # refresh existing pages
deepdoc generate --clean   # delete owned files and saved state, then rebuild

Flag	Description
`--force`	Fully refresh existing DeepDoc-managed docs instead of refusing
`--clean`	Delete DeepDoc-owned files and saved state, then rebuild
`--yes`	Skip confirmation prompts
`--include` / `--exclude`	Extra glob patterns for this run
`--docs` / `--site`	Use different output directories, this run only
`--batch-size`	Pages submitted per batch
`--max-parallel-workers`	Concurrent LLM calls
`--rate-limit-pause`	Seconds between batches
`--strict-quality`	Fail if any page is invalid, degraded, fallback or a stub
`--deploy`	Build the static site afterwards

`update`

The normal command after your first generate. Diffs against the last synced commit and regenerates only the affected pages.

deepdoc update

Flag	Default	Description
`--since`	last synced commit	Git ref to diff against
`--replan`	off	Force a full replan even if the change set looks incremental
`--strict-quality`	off	Exit non-zero when any page fails validation — use in CI
`--deploy`	off	Deploy after a fully successful update

See Keeping docs fresh for how it decides what to redo.

`serve`

Previews the docs locally with live reload.

deepdoc serve
deepdoc serve --port 4000

Before starting, it re-applies .deepdoc.yaml from your saved plan — so config changes show up here with no regeneration and no LLM calls:

Applied .deepdoc.yaml changes (no regeneration, no LLM calls): colors, nav

Needs Node.js 18+. The first run installs npm dependencies automatically.

`deploy`

Builds a static site into <site_dir>/out/, publishable to any static host.

deepdoc deploy

Refuses to ship broken docs:

Refusing to deploy docs with unresolved quality issues:
- invalid docs present: cli-commands
- stub docs present: page-generation-validation

Fix those by regenerating. See Deploying.

`status`

Shows what has been generated and what is now stale. No flags, no cost.

deepdoc status

`config`

deepdoc config show
deepdoc config set site.theme.preset ocean
deepdoc config set site.colors.primary "#7c3aed"

config set rejects keys that do not exist, rather than silently writing a setting that does nothing:

Error: Unknown config key: site.theme.presset

Did you mean:
  site.theme.preset

`performance`

Timing, token usage, retries and I/O for recent runs, from .deepdoc/performance/runs.jsonl.

deepdoc performance

This is where to look if you want to know what a run actually cost.

`clean`

Resets the repository to a pre-DeepDoc state.

deepdoc clean          # prompts first
deepdoc clean --yes    # no prompt

Removes only files DeepDoc created — tracked against its own manifest — so hand-written pages in your docs directory and chatbot_backend/ survive.

`benchmark`

Scores planner quality against a gold manifest. Mainly of interest if you are working on DeepDoc itself.

deepdoc benchmark --catalog benchmarks/catalog.json
deepdoc benchmark --generated-root ./deepdoc-docs --scorecard-out .deepdoc/scorecards/latest.json

Use --strict-scorecard to fail when completeness gates are not met. Without a hand-written catalog, artifact mode (--generated-root) computes a provisional scorecard from persisted .deepdoc/ output.

Keeping docs fresh

How update decides what to regenerate.

What it costs

Which commands spend money, and how much.