OpenAI-compatible troubleshooting

Five checks before running DeepSeek, Qwen, or Kimi through an OpenAI-compatible API.

For overseas developers testing Chinese-strong models in coding assistants, document pipelines, localization QA, and batch jobs, compatibility should be verified before production traffic moves.

Create account Read the API overview

Start outside the app

Run one tiny chat request before debugging framework wrappers, agents, or coding assistants.

Verify feature flags

Check JSON output, tools, streaming, and provider-specific fields separately.

Keep the data boundary

Use China-model routes only for non-sensitive, retryable workloads that fit your policy.

1. Base URL and auth smoke test

Most integration bugs are easier to isolate with a raw request. Confirm the base URL path, API key, model ID, and whether your SDK expects the base URL to include /v1.

curl "$OPENAI_BASE_URL/chat/completions" \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"YOUR_MODEL_ID","messages":[{"role":"user","content":"Reply with one short sentence."}],"stream":false}'

2. OpenAI-compatible is request-shape compatible

A provider may accept the familiar chat completions shape without supporting every OpenAI feature exactly the same way. Build a small capability matrix for:

response_format, JSON mode, and strict schema behavior
tools, tool_choice, and malformed tool-output handling
streaming deltas and reconnect/retry behavior
sampling parameters such as top_p, stop, seed, and penalties

3. Add fallbacks for structured output

If normal chat works but response_format fails, the route is reachable. Treat the provider as chat-compatible, fall back to prompt-for-JSON plus validation, and record a clear unsupported-parameter error.

4. Preserve provider-specific fields

Some Chinese-model routes can expose extra reasoning or metadata fields. Preserve unknown fields in sanitized debug metadata, then normalize the assistant message your application consumes.

5. Choose Botat only where it fits

Botat is an OpenAI-compatible gateway for overseas developers testing DeepSeek, Qwen, Kimi, and other Chinese-strong models for non-sensitive Chinese content, batch processing, coding assistance, and retryable tasks.

Do not send sensitive personal data, confidential customer data, regulated data, or workloads with strict residency requirements through China-model routes.