Per-agent control
Support can run a stronger model while a simple router stays light and fast.
How it works
Start with the job — receptionist, support, lead qualifier, or a focused subagent.
Pick the provider and model that fit that job's difficulty and latency needs.
If answers are thin, step up. If costs climb on simple FAQs, step down.
The main agent and its subagents can use different models for different work.
Capabilities
You should not need an engineering team to change brains.
Support can run a stronger model while a simple router stays light and fast.
A knowledge search helper and a summarizer do not have to share the same model.
Pair model choice with Embage's action-based pricing so short FAQs stay cheap.
Use a stronger model for long investigations, refunds, or ambiguous questions.
Change the model in settings — keep knowledge, datastores, and tools as they are.
Model choice is a field in the builder, not a deployment pipeline.
In practice
Start practical, then adjust from real conversations.
Prefer faster, lighter models when the job is greet, classify, and hand off.
Prefer stronger models when policy nuance and multi-step troubleshooting matter.
Post-conversation reviewers often want a model that follows field shapes carefully.
If most chats are short and knowledge-backed, keep the main model lean.
FAQ
No. You can change models as your traffic, quality bar, or budget changes — without rebuilding the agent.
Yes. That is one of the main reasons to split work into helpers — each job gets the brain it deserves.
Embage pricing stays close to model usage and completed actions, so choosing a lighter model for simple work shows up in what you spend.