Trust & data flow

The question we hear most is whether WorkspaceGPT stores your organisation’s data. It does not. This page says what is sent where, and shows how to confirm it without taking our word for it.

In short

  • Your Knowledge is indexed on your machine. Confluence, Jira and Azure DevOps content is synced, embedded and searched locally, and the index is stored in local files. Source code is read in place and not indexed.
  • Only the question and the snippets retrieved for it ever leave, and only to the model you chose. Your index, repository and Knowledge corpus are never uploaded.
  • You can keep us out of the path entirely. Use a local model, your Copilot plan, or your own key, and no request touches a WorkspaceGPT server.
  • The app tells you where the next question goes. A label beside the model picker shows it live.

What is sent, by mode

The mode and model you pick change one thing: where the answer is generated. They never change where your content is stored.

Local model (Ollama, or a server on localhost)

Stays on device
Sent to:
Nowhere. The model runs on your machine.
Who can see it:
Nobody.

Your GitHub Copilot plan

→ GitHub Copilot
Sent to:
GitHub Copilot, under your GitHub account and terms.
Who can see it:
GitHub, and the model provider Copilot routes to. Not WorkspaceGPT.

Your own key (OpenAI, Gemini, Groq, OpenRouter, NVIDIA, custom endpoint)

→ that provider
Sent to:
Straight from your machine to that provider, under your account with them.
Who can see it:
That provider. Not WorkspaceGPT.

Remote mode (managed inference, Preview)

→ WorkspaceGPT
Sent to:
The WorkspaceGPT endpoint, then its upstream model provider (currently OpenRouter).
Who can see it:
The endpoint and the upstream provider process it in memory. We store none of it.

If your organisation already approves GitHub Copilot, using your Copilot plan adds no new vendor. Copilot’s own retention depends on your GitHub plan, so check yours. On the desktop app the Copilot connection is a direct, unofficial path to Copilot’s API; in VS Code it goes through VS Code’s Language Model API.

Only a local model gives a guarantee that nothing leaves your machine. Every other option sends the question and retrieved snippets to the provider you picked.

What never leaves your machine

  • The vector index and embeddings (computed on-device in every mode).
  • Your repository, and your Confluence, Jira and Azure DevOps corpus.
  • Credentials for those systems. They are kept in your IDE’s secret storage, or the macOS Keychain / Windows Credential Manager on Desktop.
  • Requests to Confluence, Jira and Azure DevOps go from your machine directly to those services. They do not pass through us.

See and delete what is stored

In Settings → Knowledge, each connected source shows how many items it has stored on your device and how much space they take. Show folder opens the folder in Finder or Explorer, so you can read the files yourself. Delete local copy removes them and keeps the connection; the next sync downloads them again. Disconnecting a source deletes them too.

Check it yourself

You don’t need to trust this page. Three ways to confirm it:

  1. Watch the network. Start a network monitor (Little Snitch or Lulu on macOS, Resource Monitor or Wireshark on Windows), then sync Confluence and ask a question with a local model. You will see traffic to your Atlassian or Azure DevOps host and to PostHog analytics, and nothing carrying your content anywhere else.
  2. Read the code. The source is on GitHub. The files that matter: the Remote endpoint (it logs a status code and error type only, never a request or response body), and the analytics service.
  3. Look at what each question carries. The question and the snippets sent with it are the whole payload. Point a local proxy at your provider’s base URL (Custom provider) to read exactly what is sent.

Analytics, in full

The app sends anonymous usage events to PostHog (EU) in every mode. They carry an event name, an install identifier, versions, and the active mode, and never prompts, answers, file contents, paths or document titles. The privacy policy lists every field. There is no in-app switch for it yet; to turn it off, block eu.i.posthog.com or email us.

Remote mode, plainly

Remote mode is the one path where your question and retrieved snippets pass through a server we run. The endpoint does not log request or response bodies and writes nothing about them to a database. What we store is your GitHub identity, plan, a session token and a weekly credit count. The upstream provider is OpenRouter, and it is the authority on its own handling. We do not opt our account into prompt logging or data sharing. If you need a guarantee across the whole path, use a local model.

Security contact

Found a problem, or need our answers for a security review? Write to contact@workspacegpt.in. For a vulnerability, say so in the subject line and we will reply directly.

See the full Privacy Policy.