Add local LLM query support

README.md

@@ -56,13 +56,16 @@ MYTHIC__SERVER_PORT=7443
 MYTHIC__TIMEOUT=-1
 MYTHIC__PAYLOAD_PORT_HTTP=1337
 ```
-If you need local LLM via [ollama](https://github.com/ollama/ollama) add to env also 
+If you need local LLM via [ollama](https://github.com/ollama/ollama) add to env also this. We recommend these models [Mistral](https://ollama.com/library/mistral), [Qwen3-ab](https://ollama.com/jaahas/qwen3-abliterated) , [llama3.1:8b](https://ollama.com/library/llama3.1), [llama3.1-ab:8b](https://ollama.com/mannix/llama3.1-8b-abliterated) (more freely), [Qwen2.5-coder:32b](https://ollama.com/library/qwen2.5-coder)  which depends on your resources.
 ```env
+LLMSERVICE__LOCAL=TRUE
 LLMSERVICE__API_URL=http://localhost:69228
 LLMSERVICE__API_KEY=super_secret_key
 LLMSERVICE__TIMEOUT=120
 LLMSERVICE__DEFAULT_MODEL=mistral
 ```
+Otherwise, the values will be taken from the [config](https://github.com/eogod/EAGLE/blob/main/backend/app/core/config.py), and you can see the validation there.
+
 Install dependencies
 ```bash
 # Poetry install (python3.13)

backend/app/cmd/llm_analysis.py

@@ -1,10 +1,23 @@
 """ Module for unified interface to LLM services """
+import os
+from dotenv import load_dotenv
+
 import g4f
+from ollama import Client
 from fastapi import HTTPException
-# TODO: ollama, openrouter , yandexgpt support
+
+from app.core.llm_templ import LLMTemplates
+
 # Настройка g4f
 g4f.debug.logging = True
 
+load_dotenv()
+IS_LOCAL_LLM: bool = os.getenv('LLMSERVICE__LOCAL')
+if IS_LOCAL_LLM:
+    client_ollama = Client(host=os.getenv('LLMSERVICE__API_URL'))
+else:
+    client_ollama = None
+
 
 class LLMService:
     def __init__(self):
@@ -20,43 +33,60 @@ async def query_llm(self, prompt: str, provider_name: str = None) -> str:
         """
         # TODO: support custom system prompt
         try:
-            # Если указан провайдер, используем его
-            if provider_name and provider_name in self.providers:
-                provider = self.providers[provider_name]
-                try:
-                    response = await g4f.ChatCompletion.create_async(
-                        model=g4f.models.default,
-                        messages=[{"role": "user", "content": prompt}],
-                        provider=provider
-                    )
-                    return response
-                except Exception as e:
-                    raise HTTPException(
-                        status_code=500,
-                        detail=f"Provider {provider_name} failed: {str(e)}"
-                    ) from e
-
-            # Если провайдер не указан или не найден, пробуем разные
-            for name, provider in self.providers.items():
-                try:
-                    response = await g4f.ChatCompletion.create_async(
-                        model=g4f.models.default,
-                        messages=[{"role": "user", "content": prompt}],
-                        provider=provider
-                    )
-                    if response and len(response) > 0:
-                        return response
-                except Exception as e:
-                    print(f"Provider {name} failed: {e}")
-                    continue
-
-            return "Не удалось получить ответ от ни одного провайдера"
-
+            if IS_LOCAL_LLM:
+                res: str = self._local_llm(prompt)
+            else:
+                res: str = await self._g4f_llm(prompt, provider_name)
+            return res
         except Exception as e:
             raise HTTPException(
                 status_code=500, detail=f"Ошибка при запросе к LLM: {str(e)}"
             ) from e
 
+    def _local_llm(self, prompt: str) -> str:
+        """ query with local llm ollama """
+        res: dict = client_ollama.generate(
+            model=os.getenv('LLMSERVICE__DEFAULT_MODEL'),
+            prompt=prompt, system=LLMTemplates.SYSTEM_PROMT
+        )
+        # remove think text for deepseek-r1, qwen , qwq models
+        parts_th = res.response.rsplit('</think>', 1)
+        return parts_th[-1] if len(parts_th) > 1 else res.response
+
+    async def _g4f_llm(self, prompt: str, provider_name: str) -> str:
+        """ query with g4f proxy providers like DDG """
+        # Если указан провайдер, используем его
+        if provider_name and provider_name in self.providers:
+            provider = self.providers[provider_name]
+            try:
+                response = await g4f.ChatCompletion.create_async(
+                    model=g4f.models.default,
+                    messages=[{"role": "user", "content": prompt}],
+                    provider=provider
+                )
+                return response
+            except Exception as e:
+                raise HTTPException(
+                    status_code=500,
+                    detail=f"Provider {provider_name} failed: {str(e)}"
+                ) from e
+
+        # Если провайдер не указан или не найден, пробуем разные
+        for name, provider in self.providers.items():
+            try:
+                response = await g4f.ChatCompletion.create_async(
+                    model=g4f.models.default,
+                    messages=[{"role": "user", "content": prompt}],
+                    provider=provider
+                )
+                if response and len(response) > 0:
+                    return response
+            except Exception as e:
+                print(f"Provider {name} failed: {e}")
+                continue
+
+        return "Не удалось получить ответ от ни одного провайдера"
+
 
 # Инициализация сервиса
 llm_service = LLMService()

backend/app/core/config.py

@@ -63,10 +63,11 @@ class Mythic(BaseModel):
 
 class LLMservice(BaseModel):
     """ env format like LLMSERVICE__API_URL=http... """
-    API_URL: HttpUrl = "http://localhost:69228"  # Для локального Ollama
-    API_KEY: str = None
-    TIMEOUT: int = 120
-    DEFAULT_MODEL: str = "mistral"
+    local: bool = False
+    api_url: HttpUrl = "http://localhost:69228"  # Для локального Ollama
+    api_key: str = None
+    timeout: int = 120
+    default_model: str = "mistral"
 
 
 class Settings(BaseSettings):

backend/pyproject.toml

@@ -22,6 +22,7 @@ gql = "^3.5.3"
 dotenv = "^0.9.9"
 autopep8 = "^2.3.2"
 g4f = "^0.5.8.1"
+ollama = "^0.5.3"
 
 [tool.poetry.group.dev.dependencies]
 coverage = "^7.9.1"