Auto is the default routing mode of Model Oracle AI's unified gateway, designed so that teams can stop hand-picking a model for each coding-agent request and instead let the platform choose based on cost, quality, latency, and context. Rather than being a single trained model, Auto sits on top of multiple upstream providers and their model families, with an AI decision engine that assesses task complexity and selects an appropriate target such as a GPT-5.6 variant or Kimi K3, while shifting routine work to cheaper defaults and moving requests around slow, rate-limited, or unavailable providers. The setup documentation explicitly recommends using "auto for most work" so the gateway can pick the best available model for the request.

In practical terms, Auto is reached through Model Oracle AI's OpenAI-compatible endpoint at api.modeloracle.com, which lets coding agents like OpenCode, Cursor, Codex, and Claude Code connect without exposing provider-specific credentials; the platform keeps upstream provider keys inside its own boundary. Auto also passes through a top-level reasoning_effort parameter for reasoning-capable logical models, with supported values ranging from none/low through max depending on the underlying model family, so users can dial up analytical depth when a task warrants it. Warm, cache-friendly conversations are kept on stable models to preserve prompt-cache locality, and cross-provider fallback plus team-wide usage tracking help keep spend visible as token usage climbs.