Air-gapped deployment (Ollama)

For regulated environments, Autopilot can run with local models via Ollama. This keeps reasoning on-prem while maintaining the same approval and evidence pack workflow.

Approach

Run Ollama on a host reachable from the OpenClaw gateway
Configure OpenClaw to use an Ollama model endpoint
Install Autopilot with --skip-tailscale if VPN isn’t available
Keep runtime service bound to localhost

Tip: start with a smaller local model for cost, then upgrade the model when your workflow is stable.

Install without Tailscale

sudo ./install/install.sh --skip-tailscale

Then set your local provider details in the OpenClaw config.

Still safe

Plans remain proposals
Policies guard what is allowed
Humans approve any action
Evidence packs keep auditability