Three values, and any agent works.

Clinealready in the catalogue

Cline ships us in its built-in provider list, so there is nothing to configure by hand. Open the settings, pick LLM Tech as the API provider, paste your key. The model, base URL and capabilities come with the entry.

If your Cline is older than late August 2026 and does not list us yet, use OpenAI Compatible with the three values above, or set the key in the environment:

export LLMTECH_API_KEY="your-key"

Kilo Code

Same approach: pick OpenAI Compatible, then base URL https://api.llmtech.eu/v1 and model nvidia/Qwen3.8-27B-NVFP4. Kilo builds its provider catalogue from the same upstream source as Cline, so a built-in entry is a matter of time.

Continue

In config.yaml:

models:
  - name: Qwen3.8-27B (LLM Tech)
    provider: openai
    model: nvidia/Qwen3.8-27B-NVFP4
    apiBase: https://api.llmtech.eu/v1
    apiKey: your-key
    capabilities:
      - tool_use
      - image_input
    roles:
      - chat
      - edit

Both capabilities are declared deliberately: tool calling works, and so does image input, which most listings for this model get wrong.

aider

export OPENAI_API_BASE="https://api.llmtech.eu/v1"
export OPENAI_API_KEY="your-key"

aider --model openai/nvidia/Qwen3.8-27B-NVFP4

The openai/ prefix tells aider to treat it as an OpenAI-compatible route rather than looking the name up in its own table.

Zed

In settings.json. The key itself goes into the system keychain or the environment variable LLMTECH_API_KEY, not into this file:

{
  "language_models": {
    "openai_compatible": {
      "llmtech": {
        "api_url": "https://api.llmtech.eu/v1",
        "available_models": [
          {
            "name": "nvidia/Qwen3.8-27B-NVFP4",
            "display_name": "Qwen3.8-27B (LLM Tech)",
            "max_tokens": 262144,
            "capabilities": {
              "tools": true,
              "images": true,
              "parallel_tool_calls": true,
              "prompt_cache_key": true
            }
          }
        ]
      }
    }
  }
}

Claude Code

Claude Code speaks the Anthropic API, not the OpenAI one, so it needs a translating proxy. LiteLLM is the usual choice:

# litellm_config.yaml
model_list:
  - model_name: qwen38
    litellm_params:
      model: openai/nvidia/Qwen3.8-27B-NVFP4
      api_base: https://api.llmtech.eu/v1
      api_key: os.environ/LLMTECH_API_KEY

litellm --config litellm_config.yaml --port 4000

export ANTHROPIC_BASE_URL="http://localhost:4000"
export ANTHROPIC_MODEL="qwen38"
claude

Worth the extra step for one reason: our first business customer found this endpoint through Claude Code, which read the model page and recommended it.

Open WebUI

Admin panel, Settings, Connections, add an OpenAI API connection with base URL https://api.llmtech.eu/v1 and your key. The model list is served by /v1/models, so the model appears on its own.

LibreChat

In librechat.yaml, as a custom endpoint:

endpoints:
  custom:
    - name: "LLM Tech"
      apiKey: "${LLMTECH_API_KEY}"
      baseURL: "https://api.llmtech.eu/v1"
      models:
        default: ["nvidia/Qwen3.8-27B-NVFP4"]
        fetch: true
      titleConvo: true
      modelDisplayLabel: "Qwen3.8-27B"

Anything else

Pick the provider type OpenAI Compatible and fill in the three values. The endpoint implements chat completions with streaming, usage accounting in the stream, tool calling and structured outputs, so clients generally need no special handling. GET /v1/models is public and needs no key, if your client discovers models that way.