← Docgrity for VS Code home

Docgrity for VS Code — Privacy Policy & Data Statement

Last updated: 5 September 2026

Who we are

Docgrity for VS Code is published by Ujjavala Singh ("we"). Contact: see the support details on the VS Code Marketplace listing or GitHub profile.

The short version

• The extension runs entirely inside your VS Code. We operate no servers and can never access your repositories, findings, or accounts.
• Document text is sent to exactly one place: the language model you use through VS Code's built-in language-model API — normally your own GitHub Copilot subscription, or a local model (e.g. Ollama) if you configure one. That processing happens under your agreement with that provider (GitHub, or nobody if local).
• We receive no data at all — no content, no telemetry, no analytics, no crash reports.

What data the extension processes

Data	Where it lives	Purpose
Markdown file contents from the open workspace	In memory during a scan only	scanning for duplicates / contradictions / open questions
TF-IDF term statistics	In memory during a scan only	candidate-pair selection (fully local, no AI)
Findings, evidence quotes, model + prompt version	VS Code workspaceState (local, per-workspace)	the findings view, diagnostics, issue drafting
Git author names of scanned files	VS Code workspaceState, inside findings	potential-owner inference (labelled potential)
GitHub access token	VS Code's authentication system (we never store it)	creating the GitHub issue you explicitly approved
Settings (globs, thresholds, model choice)	VS Code settings (local or your Settings Sync)	configuration

Everything above stays on your machine (or in your own VS Code Settings Sync). Clearing findings (Docgrity: Clear findings) or uninstalling the extension removes stored findings.

What leaves your machine

• Language-model requests containing the text of the documents being assessed — sent via VS Code's vscode.lm API to the model you selected (default: your GitHub Copilot subscription; optionally a fully local model, in which case nothing leaves your machine at all).
• One GitHub API request (api.github.com) when — and only when — you approve raising an issue, containing the issue title, body and labels you previewed.

There are no other network calls. No content, metadata, or usage data is ever sent to us.

What we collect

Nothing. There is no backend, no telemetry endpoint, and no account.

Permissions the extension uses

• Workspace file read — markdown files matching your include glob, to scan them.
• Language model access (vscode.lm) — VS Code prompts you for consent on first use.
• GitHub authentication (repo scope) — requested only when you raise an issue, via VS Code's standard GitHub sign-in.
• Read-only git commands (git log, git remote get-url) — for potential-owner and repo inference.

Security measures

• Document content is treated as untrusted input; all model output is schema-validated in code and rejected if invalid.
• Evidence excerpts are verified verbatim against your files before a finding is recorded.
• No action is taken without explicit human approval — issue drafts are always previewed first.
• Tokens are never logged; the Docgrity output channel never contains document content or credentials.
• Git subprocesses run without a shell, with fixed argument lists and timeouts.

Data subject rights & retention

The extension stores no personal data outside your machine. Git author names inside findings are removed when you clear findings or uninstall. Issues you create on GitHub are governed by GitHub's terms and your repository's settings, and are fully under your control.

Changes

We'll update this page when the extension's data handling changes, with the date above.