Skip to content
Docs

Intelligent Model Routing (Complexity Routing)

model: "auto" routes each request to the most appropriate upstream via on-box, CPU-only classification (task/domain/complexity/language) — no prompt content leaves the box to route. Off by default; enabled and configured per tenant (layered over a platform default), with role-scoped access: tenant admins manage their own org; super-admins manage the platform default or any specific tenant.

Enterprise-relevant properties: task-type + complexity-tier routing; multiple upstreams per tier with failover/round-robin/weighted/random; a context-length override for large inputs; per-tenant upstream providers; full auditability (route header + a persisted decisions ring buffer + analytics panel); and no schema migration (state in system_settings; model artifacts ship in the image). Reference: AgentWatch_Complexity_Routing.md.