Local-first agents with Ollama

A guide to running XR entirely offline with open-weight models — no cloud, no telemetry, no keys.

XR is designed to run entirely on your machine: CLI, dashboard, and model access all work offline.

Local inference engines — Ollama, llama.cpp, LM Studio, vLLM — are first-class provider presets. Bring your own API keys if you prefer cloud models; keys stay in your environment and are never logged.

There is no XR cloud and no mandatory control plane. Telemetry is off by default; the only outbound call the runtime makes is to a model provider you configured.

The audit log, memory, sessions, and workflow records all live in your local state directory.

You can verify this yourself: run XR with no network and finish a local task end to end.