Install Engine + extension
ContextMint ships as a VS Code extension plus a backend. The default path is local-first: Engine on your machine manages the server and Ollama; the extension connects over loopback. Enterprise teams can run a sovereign on-prem API (OpenAI-compat / vLLM + OIDC) instead — see sovereign server · deployment models. If anything fails during install, open installation support first.
Choose your deployment
Engine + extension on the same machine. Sovereign local-first path, no shared server. Follow Steps 1–3 below.
Platform team deploys API + OpenAI-compat inference (vLLM). Extension sets ContextMint.serverUrl + enterpriseMode. Requires path alignment (Remote SSH or mounted repo). Sovereign guide →
Local-first — what you need
- Windows 10+, macOS 12+, or Linux x64
- VS Code 1.85 or later
- Ollama installed (Engine can guide first-run setup)
- A code workspace you want to index
Step 1 — ContextMint Engine
Engine installs like Ollama or Docker Desktop: a menubar/tray app with a utilitarian operator window when you open it. Pick System, Light, or Dark from the header or Settings. Download the installer from GitHub Releases.
| Windows x64 | contextmint-engine-win-x64.msi |
| macOS | contextmint-engine-macos-universal.dmg |
| Linux x64 | contextmint-engine-linux-x64.AppImage |
First launch runs a short wizard: Ollama detect, start services, and a link to install the extension. Start at login is on by default — change it in Engine → Settings.
Step 2 — ContextMint extension
Install from the VS Code Marketplace, or sideload the .vsix from the same GitHub Release as Engine.
Open any workspace folder. The extension connects to localhost:8000 and registers your repo for indexing automatically.
Step 3 — First chat
- Confirm Engine tray shows server healthy (green).
- Open VS Code on your project folder.
- Press Ctrl+Shift+Alt+L or open Chat from the activity bar.
- Wait for readiness Ready (indexing may take a few minutes on large repos).
- Optional: Engine → Models → pull a vision model (moondream) if you plan to paste UI screenshots.
- Review Context Lens, then send your first question — or paste a screenshot and ask about it.
In the compose ⋯ menu, set Context to Repo for codebase questions (default), Work for planning or comms, or Hybrid when you want repo evidence plus recommendations.
Status bar shows ContextMint | Ready when Engine and the backend are healthy. Full v1.0 capability list: Features → · User documentation → · Troubleshooting →
Optional — cloud routing (BYOK)
Not required for the default workflow. If your team approves hybrid inference, enable cloud routing in extension settings and add provider API keys via ContextMint: Configure Cloud API Key. Keys are stored in VS Code secrets on your machine — never in Engine or server config. Assembled context is redacted before any cloud dispatch. Local Ollama remains the default. Details: FAQ · Trust & Security.
Enterprise — sovereign on-prem API (pilot)
For regulated teams sharing one GPU host or monorepo index. Not a substitute for Engine on every laptop — this is a platform-operated API server with OpenAI-compatible inference (typically vLLM). Do not equate “enterprise” with “Ollama tray on a shared box only.”
- Platform team deploys the API stack (Docker Compose) with Chroma, Redis, and an OpenAI-compatible inference URL (vLLM overlay common).
- Mount or clone the repository on the server host so indexing paths match what developers open in VS Code (Remote SSH is the common pattern).
- Enable server auth (AUTH__ENABLED), configure OIDC issuer/audience, and domain allowlists.
- Each developer: install the extension only, set ContextMint.serverUrl to the team HTTPS endpoint, enable ContextMint.enterpriseMode, and set ContextMint.oidcProviderId (e.g. microsoft).
- Optional: configure ContextMint.localFallbackUrl if chat should fall back to a local engine when the team server is unreachable.
Full platform guide: Sovereign server deployment. Contact us for a design-partner pilot — architecture review, path strategy, and security questionnaire support: Trust center · Design partners.
Operate from the tray
Engine is not a daily dashboard — it is recovery and lifecycle control.
- Tray — green / amber / red at a glance; Start all / Stop all
- Overview — readiness issues when something needs attention
- Server · Ollama · Indexing — drill down only when debugging
- Logs / About — diagnostics and support bundle export
Stuck?
Common fixes before you open a GitHub issue.
- Server offline → open Engine and Start all
- Chat blocked → usually offline or starting — not indexing
- Ollama offline → Limited mode; search may work, generation degraded
Need a guided pilot?
Apply as a design partner for a 30-day evaluation pilot (extendable to 90) with direct engineering support and a written summary for stakeholders. Intake opens by start of September 2026.