A provider is the company whose model does the thinking — and the API key that pays for it. Nothing in the AI section works until at least one provider with one model exists.
Setting one up #
-
Add the provider
Name it and give its API base. Some providers are available platform-wide and need no key from you; others you connect yourself. -
Add an API key
Stored encrypted. Once saved it is never shown again in full — treat the copy in your password manager as the only copy. -
Register the models you want
A model has to be registered before an agent can select it. Register only the ones you intend to use; a long list makes the agent builder harder, not better. -
Add an embedding model if you use knowledge bases
Document search needs an embedding model specifically. Without one, knowledge bases cannot ingest anything.
Choosing a model #
The trade-off is always the same three things: quality, latency and cost per message. For WhatsApp specifically, latency matters more than people expect — a customer waiting eight seconds for a reply assumes it is broken.
| Field | What it does |
|---|---|
| Small, fast models | Good for classification, routing and short factual answers grounded in a knowledge base. Usually the right default. |
| Large models | Worth it when the agent has to reason across several tools or hold a complicated sales conversation. |
| Embedding models | Not conversational. Used only to index and search documents. |
Where to go next #
Agent builder
Persona, model, temperature, tools and the WhatsApp numbers an agent answers on.
Knowledge bases
Documents an agent searches before it answers, so replies come from your material.
Audit logs
Every inference an agent made, with status, tokens, latency and the tools it called.
Usage & limits
How much of each plan limit you have consumed this cycle.