# Native API clients (llmcpp)

**Native API clients** call a model service directly from the C++ driver. This integration-test excerpt selects the OpenAI client and limits each response:

```cpp
std::vector<std::string> args{"-fllm-no-cache",
                              "-fllm-max-output-tokens=768",
                              "-fllm-backend=openai",
                              "02_failure.cpp",
                              "-o",
                              "native-openai"};
```

The test supplies a local API stand-in, credentials for that fixture, and a requested model. No external agent command is selected for the native case.

The client translates compiler tools into provider requests, dispatches returned tool calls, and feeds results back into the exchange. Anthropic has a separate native client using the same compiler-side operations. Neither route needs Python for native transport. A plausible response alone is insufficient: an implementation must reach [[llmcpp-body-submission|submission]] and compiler acceptance.

Source: [test_integration_agents.cpp](https://github.com/yanevskiv/llmcpp/blob/main/tests/test_integration_agents.cpp).