Connection and Usage

When a call fails or a conversation is interrupted, first verify three things with a minimal request: authentication, endpoint path, and model availability.

Cannot connect to the platform?

Check:

• Whether the API Key is configured correctly, without extra spaces or newlines.
• Whether the Base URL is correct and does not contain duplicate paths.
• Whether the endpoint path matches the client protocol, such as the OpenAI-compatible path for OpenAI clients.
• Whether local network, company proxy, firewall, DNS, or HTTPS inspection blocks requests.
• Whether the API Key expired, was deleted, or belongs to a restricted account.

For OpenAI-compatible calls, test with a minimal non-streaming request:

curl https://api.requestbeam.com/openai/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-5.4",
    "messages": [
      { "role": "user", "content": "Reply with OK only." }
    ],
    "stream": false
  }'

If the minimal request fails, the issue is usually in networking, authentication, path configuration, model permissions, or account status.

Model selected but no response or an error?

Common causes:

• The model ID is incorrect or does not match the platform model list.
• The selected model is temporarily unavailable or not allowed for the API Key.
• The request format is invalid, such as wrong field names, nesting, or types.
• The account has insufficient balance, the context is too long, output is too long, or the service timed out.
• The third-party client does not handle streaming or error responses correctly.

Copy the exact model ID from the console or model list, test with a minimal body, then add context and parameters back gradually.

What if a chat errors or disconnects?

Identify where it fails:

• Before request start: check authentication, path, model, and parameters.
• During streaming: check client timeout, proxy chain, and SSE parsing.
• With long context or long output: check request size, output limits, and service timeout.

Recommended handling:

1. Read the specific error fields.
2. Temporarily switch to non-streaming to distinguish model failure from streaming parsing failure.
3. Add backoff retry for 429, network instability, and some 5xx errors.
4. Include required context on retries instead of assuming server-side conversation state.
5. Record X-Request-ID if the problem persists.