Three platforms. No account.

Verify any installer against SHA256SUMS.txt , published alongside the assets in every GitHub release . In-app updates are strictly user-triggered; the app never phones home.

Install Ollama , pull a model that fits your hardware, start the daemon. Team-X auto-detects the local instance the next time you open the app and uses it for any role configured for the Local privacy tier.

# macOS / Linux: install Ollama
curl -fsSL https://ollama.com/install.sh | sh

# Windows: download from ollama.com/download

# Pull a model. Pick one:
ollama pull llama3.1:8b     # 4.7 GB · solid generalist
ollama pull qwen2.5:14b     # 9 GB · stronger reasoning
ollama pull phi3:14b        # 7.9 GB · fast, light

# Start the daemon
ollama serve

Node 22+ and pnpm 9+ required. The clone is ~80 MB; install pulls electron-rebuild and the native deps; first build takes ~3-4 minutes; HMR is instant after that.

git clone https://github.com/git-rocky-stack/team-x.git
cd team-x
pnpm install
pnpm dev

# Run the unit suite (3,521 tests)
pnpm test

# Run the Playwright E2E suite
pnpm -F @team-x/desktop test:e2e

# Build a platform installer
pnpm dist:win        # NSIS, x64 + arm64
pnpm dist:mac        # DMG, Intel + Apple Silicon
pnpm dist:linux      # AppImage + .deb

See CONTRIBUTING.md for the full development guide, the role-pack contribution flow, and the PR rubric.

Team-X supports Anthropic, OpenAI, Google, Groq, OpenRouter, Together, Fireworks, Ollama (local), LM Studio (local), and any OpenAI-compatible endpoint. Add a provider in Settings → Providers: paste the API key, click Test, toggle on. Keys live in the OS keychain via keytar; they never enter a config file.

Privacy tier filtering is enforced at the router. Set a role to the Local tier and it physically cannot make a proprietary cloud call, regardless of what model the agent's spec says it prefers. Read more in the privacy posture .