Admin track

AI behavior

Models — choose the model for each job (text chat, voice, Canvas generation, and conversation titles) with provider, model id, temperature, max tokens and cost visibility. Start conservative and adjust from what you see in Token Usage.

System Prompt — the instructions that define your assistant's persona and rules, with variables and full version history so you can restore a previous version. Agent Config exposes the deeper session/memory/tool-gating settings for advanced tuning.

Screenshot — Models. Per-job model selection with cost in view. 🖼 Screenshot — System Prompt. Persona and rules, versioned so you can always roll back.

🖼

⚠️ Config is published, not auto-saved. Editing the system prompt, models or settings changes a draft. It reaches your live instance only when you publish. Always sanity-check in the Playground first.