feat: add native LLaMA.cpp local provider support

backend/app/model/model_platform.py

@@ -20,6 +20,7 @@
     "z.ai": "openai-compatible-model",
     "ModelArk": "openai-compatible-model",
     "grok": "openai-compatible-model",
+    "llama.cpp": "openai-compatible-model",
 }
 
 

backend/tests/app/model/test_model_platform.py

@@ -26,6 +26,7 @@ def test_normalize_model_platform_maps_known_aliases():
     assert normalize_model_platform("grok") == "openai-compatible-model"
     assert normalize_model_platform("z.ai") == "openai-compatible-model"
     assert normalize_model_platform("ModelArk") == "openai-compatible-model"
+    assert normalize_model_platform("llama.cpp") == "openai-compatible-model"
 
 
 def test_normalize_model_platform_keeps_non_alias_unchanged():
@@ -43,7 +44,7 @@ class _Model(BaseModel):
         optional_model_platform: NormalizedOptionalModelPlatform = None
 
     item = _Model(
-        model_platform="grok",
+        model_platform="llama.cpp",
         optional_model_platform="ModelArk",
     )
 

docs/core/models/local-model.md

@@ -38,6 +38,11 @@ print(f"Server started on http://localhost:{port}")
 ollama pull qwen2.5:7b
 ```
 
+```bash
+# LLaMA.cpp server https://github.com/ggml-org/llama.cpp/tree/master/tools/server
+./llama-server -m /path/to/model.gguf --host 0.0.0.0 --port 8080
+```
+
 2. Setting your model
 
 ![set_local_model](/docs/images/models_local_model.png)

docs/get_started/quick_start.md

@@ -173,7 +173,7 @@ Eigent can run in two modes. Your choice here affects how you are billed and wha
 - **Cloud Version:** We provide pre-configured, state-of-the-art models, including GPT-4.1, GPT-4.1 mini and Gemini 2.5 Pro. Using these models is the easiest way to get started and will be billed to your account based on usage (credits).
 - **Self-hosted Version:** You can connect your own models.
   - **Cloud Models:** Connect your personal accounts from providers like OpenAI, Anthropic, Qwen, Deepseek and Azure by entering your own API key.
-  - **Local Models:** For advanced users, you can run models locally using Ollama, vLLM, or SGLang.
+  - **Local Models:** For advanced users, you can run models locally using Ollama, vLLM, SGLang, LM Studio, or LLaMA.cpp server.
 
 ### **MCP Servers**