diff --git a/oracles/src/domain/llm/generate_response_use_case.py b/oracles/src/domain/llm/generate_response_use_case.py
index a6c9c39..60b1ae6 100644
--- a/oracles/src/domain/llm/generate_response_use_case.py
+++ b/oracles/src/domain/llm/generate_response_use_case.py
@@ -2,10 +2,10 @@
 from typing import List
 from typing import Optional
 
+import groq
 from groq import AsyncGroq
+import openai
 from openai import AsyncOpenAI
-from openai import RateLimitError
-from openai import APIError
 from openai.types.chat import ChatCompletion
 from groq.types.chat import ChatCompletion as GroqChatCompletion
 
@@ -16,7 +16,9 @@
 from src.entities import PromptType
 
 
-@backoff.on_exception(backoff.expo, RateLimitError)
+@backoff.on_exception(
+    backoff.expo, (openai.RateLimitError, openai.APITimeoutError), max_tries=3
+)
 async def _generate(model: str, messages: List[dict]) -> Optional[str]:
     client = AsyncOpenAI(
         api_key=settings.OPEN_AI_API_KEY,
@@ -28,7 +30,9 @@ async def _generate(model: str, messages: List[dict]) -> Optional[str]:
     return chat_completion.choices[0].message.content
 
 
-@backoff.on_exception(backoff.expo, RateLimitError)
+@backoff.on_exception(
+    backoff.expo, (openai.RateLimitError, openai.APITimeoutError), max_tries=3
+)
 async def _generate_openai_with_params(chat: Chat) -> Optional[ChatCompletion]:
     client = AsyncOpenAI(
         api_key=settings.OPEN_AI_API_KEY,
@@ -49,11 +53,16 @@ async def _generate_openai_with_params(chat: Chat) -> Optional[ChatCompletion]:
         tool_choice=chat.config.tool_choice,
         user=chat.config.user,
     )
-    assert chat_completion.choices[0].message.content or chat_completion.choices[0].message.tool_calls
+    assert (
+        chat_completion.choices[0].message.content
+        or chat_completion.choices[0].message.tool_calls
+    )
     return chat_completion
 
 
-@backoff.on_exception(backoff.expo, RateLimitError)
+@backoff.on_exception(
+    backoff.expo, (groq.RateLimitError, groq.APITimeoutError), max_tries=3
+)
 async def _generate_groq_with_params(chat: Chat) -> Optional[GroqChatCompletion]:
     client = AsyncGroq(
         api_key=settings.GROQ_API_KEY,
@@ -72,7 +81,10 @@ async def _generate_groq_with_params(chat: Chat) -> Optional[GroqChatCompletion]
         top_p=chat.config.top_p,
         user=chat.config.user,
     )
-    assert chat_completion.choices[0].message.content or chat_completion.choices[0].message.tool_calls
+    assert (
+        chat_completion.choices[0].message.content
+        or chat_completion.choices[0].message.tool_calls
+    )
     return chat_completion
 
 
@@ -80,8 +92,7 @@ async def execute(model: str, chat: Chat) -> LLMResult:
     try:
         if not chat.config or chat.prompt_type == PromptType.DEFAULT:
             chat.prompt_type = PromptType.DEFAULT
-            response = await _generate(
-                model=model, messages=chat.messages)
+            response = await _generate(model=model, messages=chat.messages)
         elif chat.prompt_type == PromptType.OPENAI:
             response = await _generate_openai_with_params(chat)
         elif chat.prompt_type == PromptType.GROQ:
@@ -92,7 +103,7 @@ async def execute(model: str, chat: Chat) -> LLMResult:
             chat_completion=response,
             error="",
         )
-    except APIError as api_error:
+    except openai.APIError as api_error:
         print(f"OpenAI API error: {api_error}", flush=True)
         return LLMResult(
             chat_completion=None,
diff --git a/oracles/src/domain/tools/image_generation/generate_image_use_case.py b/oracles/src/domain/tools/image_generation/generate_image_use_case.py
index 08df43b..3ed3bf1 100644
--- a/oracles/src/domain/tools/image_generation/generate_image_use_case.py
+++ b/oracles/src/domain/tools/image_generation/generate_image_use_case.py
@@ -1,13 +1,14 @@
 import backoff
 import settings
 from typing import Optional
+import openai
 from openai import AsyncOpenAI
-from openai import RateLimitError
-from openai import APIError
 from src.domain.tools.image_generation.entities import ImageGenerationResult
 
 
-@backoff.on_exception(backoff.expo, RateLimitError)
+@backoff.on_exception(
+    backoff.expo, (openai.RateLimitError, openai.APITimeoutError), max_tries=3
+)
 async def _generate_image(prompt: str) -> Optional[ImageGenerationResult]:
     client = AsyncOpenAI(
         api_key=settings.OPEN_AI_API_KEY,
@@ -29,7 +30,7 @@ async def execute(prompt: str) -> Optional[ImageGenerationResult]:
             url=response.data[0].url,
             error="",
         )
-    except APIError as api_error:
+    except openai.APIError as api_error:
         print(f"OpenAI API error: {api_error}", flush=True)
         return ImageGenerationResult(
             url="",
diff --git a/oracles/src/repositories/knowledge_base_repository.py b/oracles/src/repositories/knowledge_base_repository.py
index 53c20a7..c4125e8 100644
--- a/oracles/src/repositories/knowledge_base_repository.py
+++ b/oracles/src/repositories/knowledge_base_repository.py
@@ -5,8 +5,8 @@
 import settings
 import numpy as np
 from io import BytesIO
+import openai
 from openai import AsyncOpenAI
-from openai import RateLimitError
 from collections import OrderedDict
 from typing import List, Any, Tuple, Dict
 from src.domain.knowledge_base.entities import Document
@@ -92,7 +92,9 @@ async def exists(self, name: str) -> bool:
         except KeyError:
             return False
 
-    @backoff.on_exception(backoff.expo, RateLimitError, max_tries=3)
+    @backoff.on_exception(
+        backoff.expo, (openai.RateLimitError, openai.APITimeoutError), max_tries=3
+    )
     async def _create_embedding(self, texts: List[str]) -> List[float]:
         response = await self.openai_client.embeddings.create(
             input=texts, model="text-embedding-3-small"