# gratis Choose `grts/auto` to route through authenticated native chat models with explicitly free IDs and zero catalog prices. ## Use Install through [Installation](../../README.md#install), rebuild, restart Pi, then choose `grts/auto` in `/model`. Pi owns provider authentication through its normal provider login and credential flow. Physical provider models remain selectable through their native provider IDs. Gratis accepts only available native models with zero input, output, cache-read, and cache-write prices. Eligible IDs are OpenRouter `:free` or `openrouter/free`, Google `gemma-*`, Mistral `labs-devstral-small-2512` or `labs-leanstral-1-5-1`, and zero-priced NVIDIA chat models. OpenRouter `auto`, Z.ai Coding Plan, arbitrary custom, faux, virtual, unknown-price, and positive-price models are excluded. A user request selects the first eligible model, preferring OpenRouter's free pool. The candidate list alternates available providers before trying more models from one provider. Continuations stay on the previous eligible physical model. Only a Pi-issued retry can advance to another candidate, from a pool of at most eight models. Non-retryable failures or disabled Pi retries stop without failover. No eligible model produces a clear error. Exhausted Pi retries produce a clear error. Direct requests select normally without retaining router state. `GRATIS_*` credentials, `grts//` pins, `/gratis`, network discovery, and in-stream failover are not supported. Catalog prices and free-tier availability can change; Gratis cannot guarantee provider billing or remaining quota. Free providers may retain prompts or use them for training. Do not send sensitive content unless that provider's terms are acceptable.