Models & routing

Run GrootDesk on the model you trust, and let smart routing keep cost low without sacrificing quality.

Bring your own model

GrootDesk is model-agnostic. Use your own key or our managed models:

Smart routing

The unified classifier estimates each message's complexity and routes accordingly — a fast, inexpensive model for simple questions, a stronger model for complex or sensitive ones (and for frustrated customers). You get quality where it matters and savings everywhere else.

• Simple FAQ / status → light model
• Nuanced, multi-step or escalation-prone → strong model
• Deterministic actions → no model call at all

Streaming

Replies stream back token-by-token over WebSocket, so customers see an answer forming immediately instead of waiting for a full response.

Cost control

Between smart routing, a semantic response cache and deterministic instant-actions, most conversations cost a fraction of a naïve single-model setup. The dashboard shows per-conversation spend so there are no surprises. See the architecture reference for the full picture.