LLM-Bridge 1.8.1__tar.gz → 1.9.0__tar.gz

{llm_bridge-1.8.1 → llm_bridge-1.9.0/LLM_Bridge.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: LLM-Bridge
-Version: 1.8.1
+Version: 1.9.0
 Summary: A Bridge for LLMs
 Author-email: windsnow1025 <windsnow1025@gmail.com>
 License-Expression: MIT
@@ -13,10 +13,10 @@ License-File: LICENSE
 Requires-Dist: fastapi
 Requires-Dist: httpx
 Requires-Dist: tenacity
-Requires-Dist: openai==1.99.6
+Requires-Dist: openai==1.106.1
 Requires-Dist: tiktoken==0.11.0
 Requires-Dist: google-genai==1.28.0
-Requires-Dist: anthropic==0.62.0
+Requires-Dist: anthropic==0.66.0
 Requires-Dist: PyMuPDF
 Requires-Dist: docxlatex>=1.1.1
 Requires-Dist: openpyxl
@@ -49,14 +49,20 @@ PyPI: [https://pypi.org/project/LLM-Bridge/](https://pypi.org/project/LLM-Bridge
 
 ### Model Features
 
-The features listed represent the maximum capabilities of each API type, not necessarily those of every individual model.
+The features listed represent the maximum capabilities of each API type supported by LLM Bridge.
 
-| Model Type | Input Format                   | Capabilities         | Output Format |
-|------------|--------------------------------|----------------------|---------------|
-| OpenAI     | Text, Image                    | Thinking, Web Search | Text          |
-| Gemini     | Text, Image, Video, Audio, PDF | Thinking, Web Search | Text, Image   |
-| Claude     | Text, Image, PDF               | Thinking, Web Search | Text          |
-| Grok       | Text, Image                    |                      | Text          |
+| API Type | Input Format                   | Capabilities                                               | Output Format |
+|----------|--------------------------------|------------------------------------------------------------|---------------|
+| OpenAI   | Text, Image                    | Thinking, Web Search, Code Execution                       | Text          |
+| Gemini   | Text, Image, Video, Audio, PDF | Thinking + Thought, Web Search + Citations, Code Execution | Text, Image   |
+| Claude   | Text, Image, PDF               | Thinking, Web Search                                       | Text          |
+| Grok     | Text, Image                    |                                                            | Text          |
+
+#### Planned Features
+
+- OpenAI: Web Search: Citations, Image Output
+- Gemini: Code Execution: Code, Code Output
+- Claude: Code Execution, File Output
 
 ## Installation
 

{llm_bridge-1.8.1 → llm_bridge-1.9.0}/LLM_Bridge.egg-info/requires.txt

@@ -1,10 +1,10 @@
 fastapi
 httpx
 tenacity
-openai==1.99.6
+openai==1.106.1
 tiktoken==0.11.0
 google-genai==1.28.0
-anthropic==0.62.0
+anthropic==0.66.0
 PyMuPDF
 docxlatex>=1.1.1
 openpyxl

{llm_bridge-1.8.1/LLM_Bridge.egg-info → llm_bridge-1.9.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: LLM-Bridge
-Version: 1.8.1
+Version: 1.9.0
 Summary: A Bridge for LLMs
 Author-email: windsnow1025 <windsnow1025@gmail.com>
 License-Expression: MIT
@@ -13,10 +13,10 @@ License-File: LICENSE
 Requires-Dist: fastapi
 Requires-Dist: httpx
 Requires-Dist: tenacity
-Requires-Dist: openai==1.99.6
+Requires-Dist: openai==1.106.1
 Requires-Dist: tiktoken==0.11.0
 Requires-Dist: google-genai==1.28.0
-Requires-Dist: anthropic==0.62.0
+Requires-Dist: anthropic==0.66.0
 Requires-Dist: PyMuPDF
 Requires-Dist: docxlatex>=1.1.1
 Requires-Dist: openpyxl
@@ -49,14 +49,20 @@ PyPI: [https://pypi.org/project/LLM-Bridge/](https://pypi.org/project/LLM-Bridge
 
 ### Model Features
 
-The features listed represent the maximum capabilities of each API type, not necessarily those of every individual model.
+The features listed represent the maximum capabilities of each API type supported by LLM Bridge.
 
-| Model Type | Input Format                   | Capabilities         | Output Format |
-|------------|--------------------------------|----------------------|---------------|
-| OpenAI     | Text, Image                    | Thinking, Web Search | Text          |
-| Gemini     | Text, Image, Video, Audio, PDF | Thinking, Web Search | Text, Image   |
-| Claude     | Text, Image, PDF               | Thinking, Web Search | Text          |
-| Grok       | Text, Image                    |                      | Text          |
+| API Type | Input Format                   | Capabilities                                               | Output Format |
+|----------|--------------------------------|------------------------------------------------------------|---------------|
+| OpenAI   | Text, Image                    | Thinking, Web Search, Code Execution                       | Text          |
+| Gemini   | Text, Image, Video, Audio, PDF | Thinking + Thought, Web Search + Citations, Code Execution | Text, Image   |
+| Claude   | Text, Image, PDF               | Thinking, Web Search                                       | Text          |
+| Grok     | Text, Image                    |                                                            | Text          |
+
+#### Planned Features
+
+- OpenAI: Web Search: Citations, Image Output
+- Gemini: Code Execution: Code, Code Output
+- Claude: Code Execution, File Output
 
 ## Installation
 

{llm_bridge-1.8.1 → llm_bridge-1.9.0}/README.md

@@ -19,14 +19,20 @@ PyPI: [https://pypi.org/project/LLM-Bridge/](https://pypi.org/project/LLM-Bridge
 
 ### Model Features
 
-The features listed represent the maximum capabilities of each API type, not necessarily those of every individual model.
-
-| Model Type | Input Format                   | Capabilities         | Output Format |
-|------------|--------------------------------|----------------------|---------------|
-| OpenAI     | Text, Image                    | Thinking, Web Search | Text          |
-| Gemini     | Text, Image, Video, Audio, PDF | Thinking, Web Search | Text, Image   |
-| Claude     | Text, Image, PDF               | Thinking, Web Search | Text          |
-| Grok       | Text, Image                    |                      | Text          |
+The features listed represent the maximum capabilities of each API type supported by LLM Bridge.
+
+| API Type | Input Format                   | Capabilities                                               | Output Format |
+|----------|--------------------------------|------------------------------------------------------------|---------------|
+| OpenAI   | Text, Image                    | Thinking, Web Search, Code Execution                       | Text          |
+| Gemini   | Text, Image, Video, Audio, PDF | Thinking + Thought, Web Search + Citations, Code Execution | Text, Image   |
+| Claude   | Text, Image, PDF               | Thinking, Web Search                                       | Text          |
+| Grok     | Text, Image                    |                                                            | Text          |
+
+#### Planned Features
+
+- OpenAI: Web Search: Citations, Image Output
+- Gemini: Code Execution: Code, Code Output
+- Claude: Code Execution, File Output
 
 ## Installation
 

{llm_bridge-1.8.1 → llm_bridge-1.9.0}/llm_bridge/client/implementations/claude/claude_stream_response_handler.py

@@ -20,9 +20,6 @@ class ClaudeStreamResponseHandler:
                 thought += event.delta.thinking
             elif event.delta.type == "text_delta":
                 text += event.delta.text
-        elif event.type == "citation":
-            citation = event.citation
-            text += f"([{citation.title}]({citation.url})) "
 
         chat_response = ChatResponse(
             text=text,

{llm_bridge-1.8.1 → llm_bridge-1.9.0}/llm_bridge/client/implementations/claude/non_stream_claude_client.py

@@ -26,10 +26,6 @@ async def process_claude_non_stream_response(
             thought += content.thinking
         if content.type == "text":
             text += content.text
-            # Unable to test: non-streaming Claude is currently not allowed
-            if citations := content.citations:
-                for citation in citations:
-                    text += f"([{citation.title}]({citation.url})) "
 
     chat_response = ChatResponse(
         text=text,

{llm_bridge-1.8.1 → llm_bridge-1.9.0}/llm_bridge/client/implementations/gemini/gemini_response_handler.py

@@ -18,28 +18,37 @@ class GeminiResponseHandler:
             self,
             response: types.GenerateContentResponse,
     ) -> ChatResponse:
-        text = ""
-        thought = ""
-        display = None
-        image_base64 = None
-        citations = extract_citations(response)
+        text: str = ""
+        thought: str = ""
+        code: str = ""
+        code_output: str = ""
+        image: Optional[str] = None
+        display: Optional[str] = None
+        citations: list[Citation] = extract_citations(response)
         input_tokens, stage_output_tokens = await count_gemini_tokens(response)
 
         printing_status = None
         if candidates := response.candidates:
             if candidates[0].content.parts:
                 for part in response.candidates[0].content.parts:
-                    # Thought Output
-                    if part.text:
+                    if part.text is not None:
+                        # Thought
                         if part.thought:
                             printing_status = PrintingStatus.Thought
                             thought += part.text
+                        # Text
                         elif not part.thought:
                             printing_status = PrintingStatus.Response
                             text += part.text
-                    # Image Output
-                    elif part.inline_data:
-                        image_base64 = base64.b64encode(part.inline_data.data).decode('utf-8')
+                    # Code (Causing Error)
+                    # if part.executable_code is not None:
+                    #     code += part.executable_code.code
+                    # Code Output
+                    if part.code_execution_result is not None:
+                        code_output += part.code_execution_result.output
+                    # Image
+                    if part.inline_data is not None:
+                        image = base64.b64encode(part.inline_data.data).decode('utf-8')
 
         # Grounding Sources
         if candidates := response.candidates:
@@ -63,7 +72,9 @@ class GeminiResponseHandler:
         return ChatResponse(
             text=text,
             thought=thought,
-            image=image_base64,
+            code=code,
+            code_output=code_output,
+            image=image,
             display=display,
             citations=citations,
             input_tokens=input_tokens,
@@ -80,4 +91,4 @@ def extract_citations(response: types.GenerateContentResponse) -> list[Citation]
                     citation_indices = [index + 1 for index in grounding_support.grounding_chunk_indices]
                     citation_text = grounding_support.segment.text
                     citations.append(Citation(text=citation_text, indices=citation_indices))
-    return citations
+    return citations

{llm_bridge-1.8.1 → llm_bridge-1.9.0}/llm_bridge/client/implementations/gemini/gemini_token_counter.py

@@ -5,6 +5,8 @@ async def count_gemini_tokens(
         response: types.GenerateContentResponse
 ) -> tuple[int, int]:
     usage_metadata = response.usage_metadata
+    if usage_metadata is None:
+        return 0, 0
     input_tokens = usage_metadata.prompt_token_count
     output_tokens = usage_metadata.candidates_token_count
     if output_tokens is None:

{llm_bridge-1.8.1 → llm_bridge-1.9.0}/llm_bridge/client/implementations/gemini/stream_gemini_client.py

@@ -4,7 +4,6 @@ from typing import AsyncGenerator
 
 import httpx
 from fastapi import HTTPException
-from google.genai import types
 
 from llm_bridge.client.implementations.gemini.gemini_response_handler import GeminiResponseHandler
 from llm_bridge.client.model_client.gemini_client import GeminiClient
@@ -39,7 +38,6 @@ class StreamGeminiClient(GeminiClient):
             response_handler = GeminiResponseHandler()
             async for response_delta in response:
                 yield await response_handler.process_gemini_response(response_delta)
-
         except Exception as e:
             logging.exception(e)
             yield ChatResponse(error=repr(e))

{llm_bridge-1.8.1 → llm_bridge-1.9.0}/llm_bridge/client/implementations/openai/non_stream_openai_responses_client.py

@@ -1,5 +1,7 @@
 import logging
 import re
+from pprint import pprint
+from typing import Optional
 
 import httpx
 import openai
@@ -21,7 +23,8 @@ def process_openai_responses_non_stream_response(
 
     output_list = response.output
 
-    text = ""
+    text: str = ""
+    image: Optional[str] = None
     citations: list[Citation] = []
 
     for output in output_list:
@@ -29,15 +32,25 @@ def process_openai_responses_non_stream_response(
             for content in output.content:
                 if content.type == "output_text":
                     text += content.text
-                # Citation is currently not working well in OpenAI Responses API
-                if annotations := content.annotations:
-                    for annotation in annotations:
-                        text = content.text[annotation.start_index:annotation.end_index]
+                # Citation is unavailable in OpenAI Responses API
+                # if annotations := content.annotations:
+                #     for annotation in annotations:
+                #         citations.append(
+                #             Citation(
+                #                 text=content.text[annotation.start_index:annotation.end_index],
+                #                 url=annotation.url
+                #             )
+                #         )
+        # Image Generation untestable due to organization verification requirement
+        # if output.type == "image_generation_call":
+        #     image = output.result
 
-    chat_response = ChatResponse(text=text, citations=citations)
+    chat_response = ChatResponse(text=text, image=image)
     output_tokens = count_openai_output_tokens(chat_response)
     return ChatResponse(
         text=text,
+        image=image,
+        citations=citations,
         input_tokens=input_tokens,
         output_tokens=output_tokens,
     )
@@ -54,6 +67,7 @@ class NonStreamOpenAIResponsesClient(OpenAIClient):
 
             response: Response = await self.client.responses.create(
                 model=self.model,
+                reasoning=self.reasoning,
                 input=serialize(self.messages),
                 temperature=self.temperature,
                 stream=False,

{llm_bridge-1.8.1 → llm_bridge-1.9.0}/llm_bridge/client/implementations/openai/steam_openai_responses_client.py

@@ -1,6 +1,7 @@
 import logging
 import re
-from typing import AsyncGenerator
+from pprint import pprint
+from typing import AsyncGenerator, Optional
 
 import httpx
 import openai
@@ -11,16 +12,30 @@ from openai.types.responses import ResponseStreamEvent
 from llm_bridge.client.implementations.openai.openai_token_couter import count_openai_responses_input_tokens, \
     count_openai_output_tokens
 from llm_bridge.client.model_client.openai_client import OpenAIClient
-from llm_bridge.type.chat_response import ChatResponse
+from llm_bridge.type.chat_response import ChatResponse, Citation
 from llm_bridge.type.serializer import serialize
 
 
-def process_delta(event: ResponseStreamEvent) -> str:
-    if event.type != "response.output_text.delta":
-        return ""
+def process_delta(event: ResponseStreamEvent) -> ChatResponse:
+    text: str = ""
+    image: Optional[str] = None
+    citations: list[Citation] = []
 
-    content_delta = event.delta
-    return content_delta
+    if event.type == "response.output_text.delta":
+        text = event.delta
+    # Citation is unavailable in OpenAI Responses API
+    if event.type == "response.output_text.annotation.added":
+        pass
+    # Image Generation untestable due to organization verification requirement
+    # if event.type == "response.image_generation_call.partial_image":
+    #     image = event.partial_image_b64
+
+    chat_response = ChatResponse(
+        text=text,
+        image=image,
+        citations=citations,
+    )
+    return chat_response
 
 
 async def generate_chunk(
@@ -29,11 +44,12 @@ async def generate_chunk(
 ) -> AsyncGenerator[ChatResponse, None]:
     try:
         async for event in stream:
-            content_delta = process_delta(event)
-            chat_response = ChatResponse(text=content_delta)
+            chat_response = process_delta(event)
             output_tokens = count_openai_output_tokens(chat_response)
             yield ChatResponse(
-                text=content_delta,
+                text=chat_response.text,
+                image=chat_response.image,
+                citations=chat_response.citations,
                 input_tokens=input_tokens,
                 output_tokens=output_tokens,
             )
@@ -53,6 +69,7 @@ class StreamOpenAIResponsesClient(OpenAIClient):
 
             stream: AsyncStream[ResponseStreamEvent] = await self.client.responses.create(
                 model=self.model,
+                reasoning=self.reasoning,
                 input=serialize(self.messages),
                 temperature=self.temperature,
                 stream=True,

{llm_bridge-1.8.1 → llm_bridge-1.9.0}/llm_bridge/client/model_client/openai_client.py

@@ -1,6 +1,7 @@
 from typing import AsyncGenerator, Iterable
 
 import openai.lib.azure
+from openai.types import Reasoning
 from openai.types.responses import ToolParam
 
 from llm_bridge.client.chat_client import ChatClient
@@ -18,6 +19,7 @@ class OpenAIClient(ChatClient):
             api_type: str,
             client: openai.AsyncOpenAI | openai.lib.azure.AsyncAzureOpenAI,
             tools: Iterable[ToolParam],
+            reasoning: Reasoning,
     ):
         self.model = model
         self.messages = messages
@@ -25,6 +27,7 @@ class OpenAIClient(ChatClient):
         self.api_type = api_type
         self.client = client
         self.tools = tools
+        self.reasoning = reasoning
 
     async def generate_non_stream_response(self) -> ChatResponse:
         raise NotImplementedError

{llm_bridge-1.8.1 → llm_bridge-1.9.0}/llm_bridge/logic/chat_generate/model_client_factory/claude_client_factory.py

@@ -1,5 +1,6 @@
 import anthropic
-from anthropic.types import ThinkingConfigEnabledParam
+from anthropic.types import ThinkingConfigEnabledParam, AnthropicBetaParam
+from anthropic.types.beta import BetaCodeExecutionTool20250825Param, BetaWebSearchTool20250305Param
 
 from llm_bridge.client.implementations.claude.claude_token_counter import count_claude_input_tokens
 from llm_bridge.client.implementations.claude.non_stream_claude_client import NonStreamClaudeClient
@@ -39,11 +40,18 @@ async def create_claude_client(
         budget_tokens=16000
     )
     temperature = 1
-    betas = ["output-128k-2025-02-19"]
-    tools = [{
-        "type": "web_search_20250305",
-        "name": "web_search",
-    }]
+    betas: list[AnthropicBetaParam] = ["output-128k-2025-02-19", "code-execution-2025-08-25"]
+    tools = [
+        BetaWebSearchTool20250305Param(
+            type="web_search_20250305",
+            name="web_search",
+        ),
+        # Code Execution is unavailable in Claude
+        # BetaCodeExecutionTool20250825Param(
+        #     type="code_execution_20250825",
+        #     name="code_execution",
+        # )
+    ]
 
     if stream:
         return StreamClaudeClient(

{llm_bridge-1.8.1 → llm_bridge-1.9.0}/llm_bridge/logic/chat_generate/model_client_factory/gemini_client_factory.py

@@ -38,6 +38,11 @@ async def create_gemini_client(
                 url_context=types.UrlContext()
             )
         )
+        tools.append(
+            types.Tool(
+                code_execution=types.ToolCodeExecution()
+            )
+        )
     if "image" not in model:
         thinking_config = types.ThinkingConfig(include_thoughts=True)
     if "image" in model:

{llm_bridge-1.8.1 → llm_bridge-1.9.0}/llm_bridge/logic/chat_generate/model_client_factory/openai_client_factory.py

@@ -2,7 +2,10 @@ import re
 
 import openai
 from fastapi import HTTPException
+from openai.types import Reasoning
 from openai.types.responses import WebSearchToolParam
+from openai.types.responses.tool_param import CodeInterpreter, CodeInterpreterContainerCodeInterpreterToolAuto, \
+    ImageGeneration
 
 from llm_bridge.client.implementations.openai.non_stream_openai_client import NonStreamOpenAIClient
 from llm_bridge.client.implementations.openai.non_stream_openai_responses_client import NonStreamOpenAIResponsesClient
@@ -54,20 +57,31 @@ async def create_openai_client(
     else:
         openai_messages = await convert_messages_to_openai(messages)
 
-    tools = [
-        WebSearchToolParam(
-            type="web_search_preview",
-            search_context_size="high",
-        )
-    ]
+    tools = []
+    reasoning = None
 
-    if re.match(r"^o\d", model):
-        tools = None
-        temperature = 1
-    if re.match(r"gpt-5.*", model):
+    if model != "gpt-5-chat-latest":
+        tools.append(
+            WebSearchToolParam(
+                type="web_search",
+                search_context_size="high",
+            )
+        )
+        tools.append(
+            CodeInterpreter(
+                type="code_interpreter",
+                container=CodeInterpreterContainerCodeInterpreterToolAuto(type="auto")
+            )
+        )
+    if re.match(r"^o\d", model) or (re.match(r"gpt-5.*", model) and model != "gpt-5-chat-latest"):
         temperature = 1
-    if model == "gpt-5-chat-latest":
-        tools = None
+    if re.match(r"gpt-5.*", model) and model != "gpt-5-chat-latest":
+        reasoning = Reasoning(effort="high")
+        tools.append(
+            ImageGeneration(
+                type="image_generation",
+            )
+        )
 
     if use_responses_api:
         if stream:
@@ -78,6 +92,7 @@ async def create_openai_client(
                 api_type=api_type,
                 client=client,
                 tools=tools,
+                reasoning=reasoning,
             )
         else:
             return NonStreamOpenAIResponsesClient(
@@ -87,6 +102,7 @@ async def create_openai_client(
                 api_type=api_type,
                 client=client,
                 tools=tools,
+                reasoning=reasoning,
             )
     else:
         if stream:
@@ -97,6 +113,7 @@ async def create_openai_client(
                 api_type=api_type,
                 client=client,
                 tools=tools,
+                reasoning=reasoning,
             )
         else:
             return NonStreamOpenAIClient(
@@ -106,4 +123,5 @@ async def create_openai_client(
                 api_type=api_type,
                 client=client,
                 tools=tools,
+                reasoning=reasoning,
             )

{llm_bridge-1.8.1 → llm_bridge-1.9.0}/llm_bridge/type/chat_response.py

@@ -8,10 +8,20 @@ class Citation:
     indices: list[int]
 
 
+# TODO: adapt to different Citation formats
+# @dataclass
+# class Citation:
+#     text: str
+#     indices: Optional[list[int]] = None
+#     url: Optional[str] = None
+
+
 @dataclass
 class ChatResponse:
     text: Optional[str] = None
     thought: Optional[str] = None
+    code: Optional[str] = None
+    code_output: Optional[str] = None
     image: Optional[str] = None
     display: Optional[str] = None
     citations: Optional[list[Citation]] = None

{llm_bridge-1.8.1 → llm_bridge-1.9.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "LLM-Bridge"
-version = "1.8.1"
+version = "1.9.0"
 authors = [
     {name = "windsnow1025", email = "windsnow1025@gmail.com"}
 ]
@@ -21,10 +21,10 @@ dependencies = [
     "fastapi",
     "httpx",
     "tenacity",
-    "openai==1.99.6",
+    "openai==1.106.1",
     "tiktoken==0.11.0",
     "google-genai==1.28.0",
-    "anthropic==0.62.0",
+    "anthropic==0.66.0",
     "PyMuPDF",
     "docxlatex>=1.1.1",
     "openpyxl",