Docgrity logo

docs · integrity · vs code

Your docs are lying
to each other. Catch them.

Docgrity finds duplicates, contradictions and open questions across your repo's markdown — with verbatim evidence — and drafts the GitHub issues to fix them.

Install Free → How it works
0 servers · 0 accounts · 0 telemetry your Copilot · local Ollama · or no AI at all npm i -g docgrity

What it finds

Three kinds of doc rot, each with verbatim evidence and a confidence score.

84%Duplicates

Docs that should be one

Near-identical documents, detected semantically or by verbatim shared blocks.

91%Contradictions

Facts that don't agree

Conflicting dates, owners and numbers — quoted side by side.

95%Open questions

Questions nobody answered

Stale TODOs and unanswered questions, with potential owners from git history.

From finding to fix — real issues, drafted for you

Right-click a finding → Raise GitHub issue. Labelled, deduplicated, evidence-backed — and nothing is posted until you approve.

GitHub issues raised by Docgrity: a duplicate finding between deployment-guide.md and release-process.md, and an open-question finding in architecture-notes.md, each labelled docgrity
Actual issues raised by Docgrity — auto-labelled, fingerprint-deduplicated, auto-closed when resolved.

Up and running in a minute

InstallSearch Docgrity in Extensions (⇧⌘X)
Pick a modelCopilot, Claude, GPT, local Ollama — or none at all
Scan⇧⌘P → Docgrity: Scan repository docs
ActReview the dashboard, raise issues — you approve every post

Terminal person? CI person?

Same engine as the extension — a read-only CLI and a GitHub Action.

$ npm i -g docgrity
$ docgrity scan --provider none --open # no-agent: no model, no keys
Corpus: 5 markdown docs; assessing 9 candidate pairs (heuristic, no LLM)
2 finding(s) across 5 docs.
  [MEDIUM] duplicate — 84% confidence — deployment-guide.md + release-process.md
  [MEDIUM] open_question — 95% confidence — architecture-notes.md

CLI — read-only by design

docgrity scan --open gives an interactive HTML dashboard. Providers: none, ollama, anthropic, openai, gemini. Never posts anything.

GitHub Action — continuous

uses: ujjavala/docgrity-vscode/action@main for scheduled scans, job-summary reports and opt-in deduplicated issues. Docs.

Why it's different

Zero infrastructure

100% inside VS Code. No servers, no accounts, no telemetry.

Evidence, always

Every finding quotes exact text, verified verbatim against your files.

AI optional

Full scans with your Copilot or Ollama — or pure-algorithm no-agent mode, fully offline.

Humans approve

Every issue is a draft first. Nothing is ever posted on its own.

Settings that matter

docgrity.mode

report-and-issue or report-only — the latter blocks issue creation entirely.

docgrity.engine

ai or no-agent — pure algorithms, no model, contradictions skipped.

docgrity.checks.*

Toggle duplicates, contradictions and open questions independently.

docgrity.include / exclude

Globs for docs to scan (default **/*.md); ignored files always excluded.

docgrity.model.*

Pick the model — or run Docgrity: Select AI model. All options.

docgrity.thresholds.*

Minimum confidence per finding type — raise for fewer, surer findings.