Reame: CPU inference server that warms up and speeds up over time

Reame is an open-source CPU-based inference server with an unusual property: it gets faster the longer it runs, rather than performing at a fixed throughput. For indie makers running LLMs without GPU budgets, a self-hosted CPU server that improves with uptime could meaningfully lower the cost floor for AI features.