Automatic model-class routing
Assign models to Efficient, Balanced and Advanced classes. A configured policy selects the class for each turn, consistently across native chat, the widget, Slack, Teams, Chatwoot and API.
Choose the AI.
Keep the policy.
Configure hosted and local models, task choices, routing classes and fallbacks. Review usage and decisions within your deployment boundary.
How it supports the work
From the question to the next step.
- 01
Match the workload
Choose supported models and capabilities for retrieval, analytics, Research, Studio and decision tasks.
- 02
Apply topic policy
Set class routing, inheritance and local-only choices appropriate to the information in scope.
- 03
Review usage
Inspect accepted decisions, fallbacks, timing and cost records before extending a rollout.
Capabilities & controls
Explore model controls.
Dedicated decision models
Use configured OpenAI Decisions, Jev/SystemOne-compatible services or supported local models for intent recognition, skill selection, document requests, document style and tool selection.
Fallbacks and gradual rollout
Confident decisions can replace a general chat-model classification call. Uncertain, slow or unavailable decisions fall back to the existing classifier. Disable tasks, compare them in shadow mode or enable them individually.
Private and topic-level choices
Inherit the platform decision model or choose one per topic. Local-only options keep supported classification tasks on private infrastructure. Remote decision models are excluded from strict-PII topics; local routes have capability requirements.
Reasoning and model management
Manage providers, feature defaults, capability-aware selectors and supported thinking levels centrally. Set reasoning effort for analytics, retrieval assistance, evaluation, Research, Studio and Tutor where the model supports it.
Model gateway and decision traces
The gateway supports 132 model providers, including xAI Grok, and local OpenAI-compatible endpoints. Inspect accepted decisions, skipped tasks, fallbacks, timing and usage, with Anthropic prompt caching and cost reporting where applicable. Performance and cost depend on configuration and workload.
Availability depends on topic permissions, platform configuration, supported models and deployment. Confirm the required workflow with representative sources during evaluation.
Evaluate model controls with your team.
Bring representative sources, a useful business question and the controls your organisation needs.