# Warming (OpenCode V2)

Session warming sends periodic model requests to preserve provider-side prompt caches or short-lived state while you pause between prompts. Disabled by default.

Enable in config:
{
  "$schema": "https://opencode.ai/config.json",
  "warming": true
}

Options: prompt, interval, duration. interval can be "auto" using 90% of model cache.ttl.

Timing: starts after non-warming request; warming requests at interval; stops after duration. New activity resets timers.

Failures logged without failing session. Costs are real provider requests.