Chat
Chat agents
The same agent brain, in text. Stream replies on your website with knowledge, tools, memory and per-message billing.
What a chat agent is
A chat agent is a Katexs agent that writes instead of speaks. It has its own prompt, model, knowledge and tools, and is deployed to the website widget or used through the chat streaming API. Voice and chat are separate products in the dashboard, so a chat agent's settings never affect your phone agents.
Engines
When you deploy a chat agent, Katexs picks how it runs:
| Engine | Used when | Notes |
|---|---|---|
| Direct | A plain chat agent with an OpenAI or Anthropic key available (yours or the platform's) | Katexs calls the model directly, adds knowledge and memory, and streams the reply |
| Flow engine | The agent is in Flow mode | Runs a published flow. Beta |
If neither is available, deploy returns "Add an OpenAI or Anthropic key in Integrations to take chat live." and the agent stays in draft.
Deployment snapshots
Deploying freezes a snapshot of the agent: prompt, model, settings, tools and flow version. Visitors always talk to the last deployed snapshot, so you can edit the agent without affecting live conversations until you deploy again.
Billing
Chat is billed per message, by default 1 credit per reply. The credit is reserved when the message is accepted and refunded if the reply fails or the visitor disconnects before it completes. See Credits and pricing.
Last updated November 2026 · Found a problem? Email support@katexs.com
