Skip to main content

Self-host Vanta offline

Run Vanta fully local — no API keys, no data leaving your machine — on a local model via Ollama.

1. Install the prerequisites​

  • git — the only required tool; the Rust kernel and Node 22 are downloaded automatically by the installer (no toolchain needed).
  • Ollama for local models
ollama pull qwen2.5:14b      # the default local model
ollama serve                  # keep it running

2. Install Vanta​

git clone https://github.com/jpoindexter/Vanta ~/vanta && cd ~/vanta
./install.sh                  # downloads the prebuilt kernel + Node, installs deps, adds `vanta` to PATH

3. Point it at Ollama​

vanta setup        # choose: ollama

This writes VANTA_PROVIDER=ollama + VANTA_MODEL=qwen2.5:14b to vanta-ts/.env. No key needed. To tune the endpoint, set VANTA_OLLAMA_URL.

4. Run it​

vanta              # interactive, fully offline
vanta doctor       # confirm: kernel up, provider reachable

What still works offline​

Files, code, shell, git, the brain/memory, skills, operator systems, scheduling — all local. What needs the network: web search/fetch, browser tools, comms (Gmail/Telegram), and any cloud model.

Keep keys out of files (optional)​

If you later add a cloud provider, fetch its key from a secret manager instead of .env:

// .vanta/settings.json
{ "api_key_helper": "op read op://vault/anthropic/key" }

See Settings & secrets.

Harden execution (optional)​

VANTA_SANDBOX=1    # OS-level isolation for shell_cmd / run_code

See Security features.