5.1 AI Assistant Not Responding

Typical Symptoms

After sending a message in an AI conversation, the send button turns gray with no output, or the console reports:

• 401 Unauthorized
• 403 Forbidden
• Connection refused

Quick Diagnosis

Go to Configuration Center → AI Engine, locate the currently active model entry, and click Test Connectivity:

• Test passes → Not a configuration issue; proceed to "Correct Configuration but Still No Response"
• Test fails → Troubleshoot based on error code:
  • 401 / 403: Re-enter your API Key
  • 404: Check if the Base URL has an extra trailing /v1
  • Timeout: Domestic networks may require a proxy or mirror
  • DNS resolution failure: Your corporate network may be blocking the domain

Troubleshooting Checklist

Check Item	Method
Valid API Key	In Configuration Center → AI Engine, find the active entry and confirm the key is non-empty, doesn't contain Chinese quotation marks, and has no extra spaces; also verify in the model provider's console that the key hasn't been revoked
Sufficient balance	Most providers offer only a few million free tokens by default—exceeding this limit results in a 402 error
Service reachable	Run in terminal: curl -i -H "Authorization: Bearer YOUR_KEY" "${BASE_URL}/v1/models"; expect a 200 OK response with a list of models
Model ID exact match	Model names must match exactly. Common mistakes: claude-sonnet (wrong) → claude-sonnet-4-20250514 (correct); gpt4o-mini (wrong) → gpt-4o-mini (correct)
Correct Provider field	The Provider determines the protocol, not the URL. Use anthropic / anthropic-compatible for Anthropic Messages (x-api-key header); all others use OpenAI Completions (Authorization: Bearer)

Correct Configuration but Still No Response

Symptom	Cause	Solution
Connection drops instantly within seconds	Extra /v1 at the end of Base URL causes duplicate path concatenation	Remove the trailing /v1 (unless you're using a fixed-path reverse proxy)
Stuck loading spinner for over 60 seconds	Model is in reasoning ("thinking") mode, but frontend isn't displaying it	Temporarily set Reasoning visibility to inline in Configuration Center to observe
<think>...</think> appears in message body	Reasoning content is being sent upstream under OpenAI-compatible protocol	Studio automatically parses this via InlineThinkingRouter; if still visible, upgrade to the latest version
Single message usage resets to zero	Agent's per-turn token limit is being hit	Upgrade to the latest RDK Studio version

About Protocol Detection

Protocol detection depends solely on the Provider field—it does not auto-switch based on whether the URL contains "anthropic". If you're using a reverse proxy to wrap Anthropic services under a path that doesn't include "anthropic", you must still set the Provider to anthropic-compatible. Otherwise, Studio will send requests using the OpenAI protocol and receive a 401.

This logic is implemented in server/agent/provider-setup.ts by the resolveProtocol() function.

Permanent Solutions

• After configuration in the Configuration Center, Studio automatically persists settings to your local config directory—no need to re-enter when switching devices
• Monitor the token usage capsule in the top-right corner of AI Dock for quota warnings
• In team environments, export configurations as JSON for one-click import by colleagues