Set a hard monthly ceiling on platform LLM spend (graph extraction on our managed model). When the cap is hit, platform extraction pauses until the next month — your memories are still stored and searchable. Bring-your-own-model tenants are never capped: you pay your provider directly.