v2.0 — Local & Remote modes

WorkspaceGPT Extension Docs

Everything you need to install, configure, and get the most out of WorkspaceGPT inside VS Code, Cursor, or Antigravity.

🔐

Privacy-First

Indexing and embeddings run on-device and the vector index stays in local files — in both modes. We retain nothing.

🤖

RAG-Powered

Retrieval-Augmented Generation over your codebase, Confluence docs, and ADO tickets.

Zero Setup

Install from the marketplace and start chatting in under 2 minutes.

Installation

Available in the VS Code and Cursor marketplaces, and on Open VSX for Antigravity.

1

Via Extensions Marketplace

Open VS Code, Cursor, or Antigravity and navigate to the Extensions view:

Ctrl+Shift+X  or  Cmd+Shift+X on macOS

Search for WorkspaceGPT and click Install.

2

Via Command Palette

Press Ctrl+P to open Quick Open and run:

bash
ext install Riteshkant.workspacegpt-extension
3

Via Marketplace Website

Visit the VS Code Marketplace page and click Install.

4

On Antigravity (or other VS Code forks)

Antigravity, Windsurf, and VSCodium install from Open VSX instead of the Microsoft Marketplace. Search WorkspaceGPT in the Extensions view, or open the Open VSX page and click Download.

Minimum VS Code version: 1.98.0. WorkspaceGPT activates automatically on startup (onStartupFinished).

Modes & Privacy

WorkspaceGPT has exactly one mode switch, under Settings → Mode. It changes where answers are generated — nothing else.

Your index never moves. Embeddings are generated on-device and the vector index is written to local files inside the extension's storage, in both modes. Switching modes never uploads anything and never requires a re-index.
🔐

Local

Everything runs on your machine: the chat model, the embeddings, the index, the retrieval.

Use Ollama for a fully offline setup, or supply your own key for OpenAI, Gemini, Groq, OpenRouter, NVIDIA, or any OpenAI-compatible endpoint. No WorkspaceGPT account needed.

Remote

Preview

We run the inference infrastructure and pick the model, so there is no provider key to buy and nothing to configure.

Sign in with GitHub once. Your question and the snippets retrieved for it are sent to our endpoint per request; your documents, code and index stay on your machine.

What each mode sends

 LocalRemote
Documents, code, work itemsNever leave your machineNever leave your machine
EmbeddingsGenerated on-deviceGenerated on-device
Vector indexLocal filesLocal files
Question + retrieved snippetsTo your chosen provider, or nowhere with OllamaTo our endpoint, then the upstream model
AccountNoneGitHub sign-in, verified per request
Model keys you supplyYours, or none with OllamaNone
Stored by WorkspaceGPTNothingNothing but your account row

Zero data retention

In Remote mode your request is held in memory only for as long as it takes to stream the answer back, then discarded. We do not write prompts, answers, or retrieved snippets to any database, any file, or any log — our servers log status codes and error types only. The entirety of what we store per account is: your GitHub id and handle, your plan and status, an opaque session token that expires in 30 days, and a count of how many requests you made today.

Generation itself is performed by an upstream model provider (currently OpenRouter) under its own policy. If you need a guarantee that covers the whole path contractually, use Local mode with Ollama — no third party is involved at all. Full detail in the privacy policy.

Remote mode is in preview. It is rolling out now, and models, limits and behaviour may change while we tune it. Local mode is generally available and unaffected. Remote mode applies a fair-use daily request limit per account; you can see today's usage under Settings → Account.

AI Providers

These apply to Local mode, where you bring your own model. In Remote mode there is no provider to choose — we run the model for you (see Modes & Privacy).

ProviderPrivacyRequires API KeyNotes
Ollama100% LocalNoDefault. Run llama3.2:1b or any local model.
OpenAICloudYesGPT-4o, GPT-4-turbo, GPT-3.5 etc.
GeminiCloudYesGoogle's Gemini Pro/Flash models.
GroqCloudYesHigh-speed inference on Llama / Mixtral.
OpenRouterCloudYesAccess 100+ models via one API key.
RequestlyCloudYesCustom API endpoint proxy integration.

Configuring Ollama (Recommended)

1

Install Ollama

Download from ollama.com and follow the installer for your OS.

2

Pull a model

bash
ollama pull llama3.2:1b

For better responses, try a larger model:

bash
ollama pull llama3.2:4b
# or
ollama pull gemma3:4b
# or
ollama pull mistral
3

Select in WorkspaceGPT

Open the WorkspaceGPT sidebar → Settings → Providers → Ollama. Your locally running models will appear automatically.

Configuring Cloud Providers

Open Settings → Providers, select your provider, and paste your API key. Keys are stored securely in VS Code's secret storage and never logged.

Codebase Exploration

WorkspaceGPT explores the folder currently open in your editor with live search, file reads, and language-server navigation.

1

Open a workspace folder

Open the repository you want to discuss in VS Code, Cursor, or Antigravity.

2

Ask about the code

Ask a question in WorkspaceGPT. The agent searches and reads relevant files as it works, and can follow symbols and references through your editor's language service.

3

Review cited findings

Use the returned file references to inspect the live workspace context behind the answer.

Note: Codebase exploration does not create embeddings or a persistent index of your workspace. Availability depends on having a folder open in the current editor window.

Embeddings & Vector Storage

Connected Confluence and Azure DevOps sources are turned into vector embeddings so WorkspaceGPT can retrieve the right context. Both halves — making the embeddings and storing them — happen entirely on your machine, in either mode.

On-device, always. Indexing is not affected by the Local/Remote mode switch. There is no cloud embedding provider and no hosted vector store to configure — and therefore no way for your content to reach us.

Embedding model

ProviderModelAPI keyBest for
Text BundledXenova/all-MiniLM-L6-v2 (384-dim)Not neededConfluence pages and ADO work items. Runs on-device; first run downloads ~200 MB.

Nothing to configure — the models ship with the extension and are used automatically.

Vector storage

💾

Local files, on this machine

Vectors are written to binary files in the extension's own storage directory. They are never uploaded, mirrored, or backed up by us.

Clear them any time with Settings → Reset or the Clear Data command.

Upgrading from an older version? Earlier releases could store vectors in a Qdrant cluster. That option is gone — indexing is on-device only now. If your install used it, WorkspaceGPT tells you once on startup and you just re-sync your sources to rebuild the index locally.

Confluence Integration

Connect your Atlassian Confluence space with one-click OAuth 2.0 authentication.

1

Open Confluence settings

In the WorkspaceGPT sidebar, navigate to Settings → Confluence Integration.

2

Sign in with Atlassian

Click Sign In. A browser window will open to Atlassian's OAuth consent screen. Sign in and grant access — no passwords are stored.

The extension spins up a short-lived local HTTP server to capture the OAuth callback securely.

3

Select a space

After authentication, your accessible Confluence sites and spaces will load. Select the spaces you want to index.

4

Start Sync

Click Start Sync. Pages are fetched, converted to Markdown, embedded, and stored locally. Progress is shown in real-time.

5

Automatic background sync

The ConfluenceSyncScheduler starts automatically on extension activation and keeps your index up-to-date in the background.

Token security

OAuth access + refresh tokens are stored in VS Code's encrypted context.secrets — never in plaintext settings.

Disconnect anytime

Go to Settings → Confluence → Disconnect to revoke access and clear all synced data.

Azure DevOps Integration

Connect Azure DevOps to chat with work items, user stories, and pull requests.

1

Generate a Personal Access Token (PAT)

In Azure DevOps, go to User Settings → Personal Access Tokens → New Token.

Grant at minimum: Work Items — Read Code — Read

2

Open ADO settings

In the WorkspaceGPT sidebar, go to Settings → Azure DevOps Integration.

3

Enter your organization URL and PAT

Provide your Azure DevOps organization URL (e.g. https://dev.azure.com/your-org) and the PAT you generated.

The PAT is stored in VS Code's context.secrets, never in globalState.

4

Select project and sync

Your ADO projects will load automatically. Select a project and click Start Sync.

Work items are embedded and indexed locally — no data is sent to third-party servers.

Auth format: The extension uses Basic base64(:PAT) (colon-prefixed PAT) as required by the Azure DevOps REST API.
Enterprise

Deployment Automation

Turn the manual release checklist — syncing feature flags, env vars, and component versions across environments — into a reviewable, one-click pipeline. The model is inspired by AWS CodePipeline: a pluggable Source feeds ordered Stages of Actions. Nothing is hardwired to one team's setup.

🧩

Pluggable

Pick a source, then add only the deploy actions your org actually uses. No provider is baked in.

🛡️

Plan → Approve → Apply

Every change is previewed as a diff you approve before anything is written. Backend changes open a PR — never an auto-merge.

🔍

Discover & select

Repos, workflows, projects, and table columns are detected from your connected accounts — choose from dropdowns, don't type IDs.

Full deployment guide

Sources, actions, connections (PAT/SSO + Vercel gotchas), the Releases workflow, environments, security, troubleshooting & FAQ.

Read the guide →

MCP Server

WorkspaceGPT ships a built-in MCP (Model Context Protocol) server for GitHub Copilot and Claude Code integration.

1

Connect the MCP Server

Open the Command Palette (Cmd/Ctrl+Shift+P) and run:

vscode command
WorkspaceGPT: Connect MCP Server
2

Use with GitHub Copilot or Claude

The MCP server is registered as a definition provider for Copilot Chat (@mcp). Once connected, Copilot and Claude can query your indexed data directly via the WorkspaceGPT context.

Status bar indicator: After connecting, a WorkspaceGPT button appears in the VS Code status bar showing MCP connection health.
Companion appPairing unavailable

Chrome Extension

A browser side-panel that lets you (or a teammate) chat with your indexed Confluence pages and Azure DevOps work items — without opening VS Code. It runs entirely in the browser, talking directly to your providers; there's no WorkspaceGPT server in between.

New pairings are paused

The companion reads your vector index directly, and that index now lives only on your machine — so there is nothing for a browser on another device to connect to. The Share to Chrome action is hidden in the current extension, and the setup steps below cannot be completed on a fresh install. Existing paired installs keep working against whatever they were configured with. We will bring this back if and when a hosted index ships; the steps are kept here for reference and for anyone already set up.

🧭

Side panel

Ask questions and read grounded answers from a panel docked in Chrome.

🔗

One share code

Connect by pasting a single code generated in VS Code — no separate setup.

🚫

No server

Calls go browser → Gemini / Qdrant / your chat model directly. Nothing is proxied.

Prerequisites

Because the browser needs cloud-reachable services, the share flow required a cloud embedding provider, a Qdrant Cloud vector store, and a chat model with an API key, all configured in VS Code first. Indexing is now on-device only (see Embeddings & Vector Storage), which is exactly why pairing is paused.

Setup (for reference)

1

Install from the Chrome Web Store

Add WorkspaceGPT for Chrome to your browser.

2

Create a share code in VS Code

In the WorkspaceGPT sidebar, open Settings → Share to Chrome and click Create share code. It's copied to your clipboard.

This card is not shown in the current extension — see the notice above.

3

Paste it into the extension

Open the Chrome side panel → Settings → paste the code → Connect. You'll see a confirmation with your Qdrant URL.

4

Ask away

Close settings and chat. The extension mirrors VS Code's retrieval, so answers stay consistent across Confluence and ADO.

Treat the share code like a password

It contains your real Qdrant, Gemini, and chat-model API keys in plain form. Anyone with it can query your data and incur API costs. Share only with people you trust.

Write creds are never shared

GitHub, Vercel, Confluence, and ADO tokens stay in VS Code secret storage and are excluded from the bundle. The share is read-only knowledge access.

Scope: the share covers Confluence and Azure DevOps knowledge. Live VS Code workspace exploration is not part of the share. Regenerate the code if you rotate your keys.

Commands & Keyboard Shortcuts

All commands are accessible from the Command Palette.

CommandShortcutDescription
WorkspaceGPT: AskCmd+Shift+R / Ctrl+Shift+ROpen & focus the WorkspaceGPT chat panel.
WorkspaceGPT: New ChatStart a fresh conversation, clearing history.
WorkspaceGPT: SettingsOpen the settings panel inside the sidebar.
WorkspaceGPT: Chat HistoryBrowse and restore previous chat sessions.
WorkspaceGPT: Connect MCP ServerRegister the MCP server for Copilot / Claude.
WorkspaceGPT: Clear All Data and CacheWipe all embeddings, state, and tokens.

Activity Bar & Title Bar Icons

New Chat

Click the + icon in the WorkspaceGPT title bar to start a new session.

⚙️

Settings

The gear icon opens provider and integration configuration.

🕐

History

Browse, restore, or delete previous chat sessions.

Reset & Clear Data

WorkspaceGPT stores all data locally in VS Code's global storage. You can wipe everything at any time.

1

Via Command Palette

Run WorkspaceGPT: Clear All Data and Cache from the Command Palette. A confirmation prompt will appear.

2

Via Settings panel

Open Settings → Reset VSCode State inside the WorkspaceGPT sidebar.

Warning: This action clears all embeddings, OAuth tokens, PATs, and chat history. It cannot be undone. Background schedulers are stopped automatically before the wipe.

Troubleshooting

Common issues and how to fix them.

Extension not loading / sidebar not showing
Ensure VS Code ≥ 1.98.0. Check the Extensions panel for error banners. Try reloading the window (Cmd/Ctrl+Shift+P → Reload Window).
Ollama models not appearing
Make sure the Ollama server is running (ollama serve). The default endpoint is http://localhost:11434. Verify at least one model is pulled: ollama list.
Codebase tools are unavailable
Open a folder or workspace in the current editor window, then start a new chat turn. WorkspaceGPT explores live files and does not build a codebase index.
Confluence sync fails
Re-authenticate: Settings → Confluence → Disconnect, then Sign In again. Ensure your Atlassian account has read access to the target space.
ADO sync not working
Verify your PAT has Work Items — Read and Code — Read scopes. Re-enter the PAT in Settings → Azure DevOps. Check the organization URL format: https://dev.azure.com/your-org.
Chat returns no results or irrelevant answers
Confirm the relevant data source is indexed (check the sync status in Settings). Try re-syncing. For better quality answers, switch to a larger Ollama model or a cloud provider.
MCP server not detected by Copilot
Run WorkspaceGPT: Connect MCP Server from the Command Palette. Reload the window after connecting.