How to use the llama.cpp server web UI

The llama.cpp server includes a browser chat interface on the same HTTP listener used by its API endpoints. Using that built-in Web UI is a quick way to confirm a loaded GGUF model can answer prompts before connecting an application, coding agent, or OpenAI-compatible client.

llama-server serves the Web UI from the same process that answers health, model-list, and chat-completion requests. A working page should show the loaded model and return generated text without a separate Open WebUI or reverse proxy.

Keep the UI bound to the loopback interface for local testing unless remote users intentionally need access. A browser session can submit prompts, upload local context files, and read model output, so exposing the listener on a shared network should be treated like exposing the API itself.

1. Start llama-server with a loaded model and the Web UI enabled.

   $ llama-server \
     --hf-repo ggml-org/gemma-3-270m-it-qat-GGUF:Q4_0 \
     --host 127.0.0.1 \
     --port 8080
   llama_server: model loaded
   llama_server: listening on http://127.0.0.1:8080

   The Web UI is enabled by default unless llama-server starts with --no-ui. Replace the --hf-repo value with the GGUF model repo, quant, or local -m path used for normal inference.
   Related: How to start the llama.cpp server
   Related: How to set llama.cpp server host and port

2. Confirm the server reports ready state.

   $ curl --silent http://127.0.0.1:8080/health
   {"status":"ok"}

   An HTTP 503 response means the model is still loading. Wait for /health to return ok before using the browser.
   Related: How to check llama.cpp server health

3. Open the Web UI URL in a browser.

   http://127.0.0.1:8080

   

   The model chip beside Send should match the model loaded by llama-server. If the server was started in router mode without a model, load a model before sending a chat prompt.

4. Enter a short prompt in the message box.
5. Click Send and wait for the assistant response.

   The response area should show generated text plus generation details such as token count, elapsed time, or tokens per second.

6. Confirm the model endpoint lists the same model shown in the browser.

   $ curl --silent http://127.0.0.1:8080/v1/models
   {
     "object": "list",
     "data": [
       {
         "id": "ggml-org/gemma-3-270m-it-qat-GGUF:Q4_0",
         "owned_by": "llamacpp"
       }
     ]
   }