Agent doesn't reply

When an agent gives no answer in the Playground, the widget or the API, the cause is almost always one of the five below. The error code (in the Playground, in the stream's error event, or on the Tracing page) tells you which.

Check these in order

1. 1

   Does the agent have an LLM key?

   Agents run only on your own provider key; there's no default one. Without a key the reply ends straight away with INTERNAL_ERROR ("An unexpected error occurred"). Open the agent's Settings tab, choose a key saved under LLM keys or paste one, and save. See Add your LLM key.

2. 2

   Is the key valid?

   LLM_AUTH_ERROR ("The agent's AI provider credentials are invalid") means the provider rejected it: wrong, revoked, or the provider account has no credit. Create a new key in the provider's console and save it on the agent. If you changed a saved key, re-save every agent that uses it.

3. 3

   Do you have credits?

   At zero, requests get HTTP 402 OUT_OF_CREDITS and the widget says the assistant is unavailable. Buy a pack under Billing. See Running out of credits.

4. 4

   Is the agent active?

   A paused agent refuses API and widget requests with 403 AGENT_INACTIVE. Set it back to Active on its Settings tab.

5. 5

   Is the provider busy?

   LLM_RATE_LIMITED or LLM_UNAVAILABLE come from your provider account, not SentientOne. Wait and retry, or raise your limits with the provider.

If something goes wrong