feat(llm): switch from Anthropic to OpenAI (gpt-4o-mini)

.env.example

@@ -1,3 +1,3 @@
-ANTHROPIC_API_KEY=sk-ant-...
-ANTHROPIC_MODEL=claude-sonnet-4-6
-ALLOWED_ORIGINS=https://sheetdoc.telecom.quest
+OPENAI_API_KEY=sk-...
+OPENAI_MODEL=gpt-4o-mini
+ALLOWED_ORIGINS=https://galymzhan.telecom.quest

backend/app/config.py

@@ -4,8 +4,8 @@ from pydantic_settings import BaseSettings, SettingsConfigDict
 class Settings(BaseSettings):
     model_config = SettingsConfigDict(env_file=".env", extra="ignore")
 
-    anthropic_api_key: str = ""
-    anthropic_model: str = "claude-sonnet-4-6"
+    openai_api_key: str = ""
+    openai_model: str = "gpt-4o-mini"
     allowed_origins: str = "*"
     max_csv_bytes: int = 5 * 1024 * 1024
     max_rows: int = 1000

backend/app/llm_client.py

@@ -3,7 +3,7 @@ import json
 from pathlib import Path
 from typing import Optional
 
-from anthropic import AsyncAnthropic, APIError
+from openai import AsyncOpenAI, OpenAIError
 
 from .config import settings
 from .errors import LlmFailed
@@ -18,20 +18,50 @@ def load_prompt(name: str) -> str:
 
 class LlmClient:
     def __init__(self) -> None:
-        self._client: Optional[AsyncAnthropic] = None
+        self._client: Optional[AsyncOpenAI] = None
 
     @property
-    def client(self) -> AsyncAnthropic:
+    def client(self) -> AsyncOpenAI:
         if self._client is None:
-            if not settings.anthropic_api_key:
+            if not settings.openai_api_key:
                 raise LlmFailed(
                     what="API-ключ не настроен.",
-                    why="ANTHROPIC_API_KEY не задан.",
+                    why="OPENAI_API_KEY не задан.",
                     next="Задайте переменную окружения и перезапустите сервис.",
                 )
-            self._client = AsyncAnthropic(api_key=settings.anthropic_api_key)
+            self._client = AsyncOpenAI(api_key=settings.openai_api_key)
         return self._client
 
+    async def _chat(
+        self,
+        system: str,
+        user: str,
+        max_tokens: int,
+        temperature: float,
+        json_mode: bool,
+    ) -> str:
+        last_err: Optional[Exception] = None
+        for attempt in range(3):
+            try:
+                kwargs = {
+                    "model": settings.openai_model,
+                    "max_tokens": max_tokens,
+                    "temperature": temperature,
+                    "messages": [
+                        {"role": "system", "content": system},
+                        {"role": "user", "content": user},
+                    ],
+                }
+                if json_mode:
+                    kwargs["response_format"] = {"type": "json_object"}
+                resp = await self.client.chat.completions.create(**kwargs)
+                content = resp.choices[0].message.content or ""
+                return content.strip()
+            except OpenAIError as e:
+                last_err = e
+                await asyncio.sleep(2 ** attempt)
+        raise LlmFailed(why=f"OpenAI API: {last_err}")
+
     async def complete(
         self,
         system: str,
@@ -39,31 +69,10 @@ class LlmClient:
         max_tokens: int = 2048,
         temperature: float = 0.4,
     ) -> str:
-        last_err: Optional[Exception] = None
-        for attempt in range(3):
-            try:
-                msg = await self.client.messages.create(
-                    model=settings.anthropic_model,
-                    max_tokens=max_tokens,
-                    temperature=temperature,
-                    system=[
-                        {
-                            "type": "text",
-                            "text": system,
-                            "cache_control": {"type": "ephemeral"},
-                        }
-                    ],
-                    messages=[{"role": "user", "content": user}],
-                )
-                parts = [b.text for b in msg.content if getattr(b, "text", None)]
-                return "".join(parts).strip()
-            except APIError as e:
-                last_err = e
-                await asyncio.sleep(2 ** attempt)
-        raise LlmFailed(why=f"Anthropic API: {last_err}")
+        return await self._chat(system, user, max_tokens, temperature, json_mode=False)
 
     async def complete_json(self, system: str, user: str, max_tokens: int = 2048) -> dict:
-        raw = await self.complete(system, user, max_tokens=max_tokens, temperature=0.2)
+        raw = await self._chat(system, user, max_tokens, temperature=0.2, json_mode=True)
         raw = raw.strip()
         if raw.startswith("```"):
             raw = raw.strip("`")

backend/app/main.py

@@ -41,7 +41,7 @@ limiter = Limiter(key_func=get_remote_address)
 @asynccontextmanager
 async def lifespan(_: FastAPI):
     setup_logging()
-    log.info("startup", model=settings.anthropic_model)
+    log.info("startup", model=settings.openai_model)
     yield
 
 

backend/requirements.txt

@@ -3,7 +3,7 @@ uvicorn[standard]==0.30.6
 httpx==0.27.2
 pydantic==2.9.2
 pydantic-settings==2.5.2
-anthropic==0.39.0
+openai==1.54.4
 python-dateutil==2.9.0
 structlog==24.4.0
 cachetools==5.5.0

docker-compose.yml

@@ -4,8 +4,8 @@ services:
     image: sheetdoc:latest
     restart: unless-stopped
     environment:
-      ANTHROPIC_API_KEY: ${ANTHROPIC_API_KEY}
-      ANTHROPIC_MODEL: ${ANTHROPIC_MODEL:-claude-sonnet-4-6}
+      OPENAI_API_KEY: ${OPENAI_API_KEY}
+      OPENAI_MODEL: ${OPENAI_MODEL:-gpt-4o-mini}
       ALLOWED_ORIGINS: ${ALLOWED_ORIGINS:-https://galymzhan.telecom.quest}
     networks:
       - traefik_default