Feat/add grok 4

README.md

@@ -76,6 +76,9 @@ You can view the full traces of some samples here: [GAIA Benchmark Traces](https
 1. Clone the repository
 2. Run the following command
 
+<img width="821" alt="Screenshot 2025-07-08 at 17 50 34" src="https://github.com/user-attachments/assets/094f73aa-7384-4500-a670-528853f92ae7" />
+
+
 ```
 chmod +x start.sh
 ./start.sh

docker/sandbox/Dockerfile

@@ -14,11 +14,14 @@ RUN apt-get update && apt-get install -y \
     unzip
 
 COPY src/ii_agent/utils/tool_client /app/ii_client
+COPY ./ /app/
 
 WORKDIR /app
 
 RUN pip install -r ii_client/requirements.txt
 
+RUN pip install .
+
 RUN curl -fsSL https://bun.sh/install | bash
 RUN curl -fsSL https://code-server.dev/install.sh | sh
 

frontend/constants/models.tsx

@@ -136,4 +136,4 @@ export const PROVIDER_MODELS: { [key: string]: IModel[] } = {
       provider: "openai",
     },
   ],
-};
+};

src/ii_agent/agents/function_call.py

@@ -38,7 +38,7 @@ class FunctionCallAgent(BaseAgent):
 A general agent that can accomplish tasks and answer questions.
 
 If you are faced with a task that involves more than a few steps, or if the task is complex, or if the instructions are very long,
-try breaking down the task into smaller steps. After call this tool to update or create a plan, use write_file or str_replace_tool to update the plan to todo.md
+try breaking down the task into smaller steps. After call this tool to update or create a plan, use Write or Edit tools to update the plan to todo.md
 """
     input_schema = {
         "type": "object",

src/ii_agent/core/config/llm_config.py

@@ -23,7 +23,7 @@ class LLMConfig(BaseModel):
     model: str = Field(default=DEFAULT_MODEL)
     api_key: SecretStr | None = Field(default=None)
     base_url: str | None = Field(default=None)
-    max_retries: int = Field(default=3)
+    max_retries: int = Field(default=5)
     max_message_chars: int = Field(default=30_000)
     temperature: float = Field(default=0.0)
     vertex_region: str | None = Field(default=None)

src/ii_agent/llm/__init__.py

@@ -7,9 +7,7 @@
 def get_client(config: LLMConfig) -> LLMClient:
     """Get a client for a given client name."""
     if config.api_type == APITypes.ANTHROPIC:
-        return AnthropicDirectClient(
-            llm_config=config,
-        )
+        return AnthropicDirectClient(llm_config=config)
     elif config.api_type == APITypes.OPENAI:
         return OpenAIDirectClient(llm_config=config)
     elif config.api_type == APITypes.GEMINI:

src/ii_agent/llm/gemini.py

@@ -127,7 +127,6 @@ def generate(
                     config=types.GenerateContentConfig(
                         tools=tool_params,
                         system_instruction=system_prompt,
-                        temperature=temperature,
                         max_output_tokens=max_tokens,
                         tool_config={'function_calling_config': {'mode': mode}}
                         ),
@@ -150,8 +149,17 @@ def generate(
                     raise e
 
         internal_messages = []
-        if response.text:
-            internal_messages.append(TextResult(text=response.text))
+        text_parts = []
+        if response.candidates and len(response.candidates) > 0:
+            candidate = response.candidates[0]
+            if candidate.content and candidate.content.parts:
+                for part in candidate.content.parts:
+                    if hasattr(part, 'text') and part.text:
+                        text_parts.append(part.text)
+
+        if text_parts:
+            combined_text = ''.join(text_parts)
+            internal_messages.append(TextResult(text=combined_text))
 
         if response.function_calls:
             for fn_call in response.function_calls:

src/ii_agent/llm/openai.py

@@ -110,9 +110,8 @@ def generate(
                 role = "user"
             elif str(type(internal_message)) == str(TextResult):
                 internal_message = cast(TextResult, internal_message)
-                # For TextResult (assistant), content is handled differently by OpenAI API
-                message_content_obj = {"type": "text", "text": internal_message.text}
-                openai_message = {"role": "assistant", "content": [message_content_obj]}
+                # For TextResult (assistant), OpenAI expects content as a string for regular messages
+                openai_message = {"role": "assistant", "content": internal_message.text}
                 openai_messages.append(openai_message)
                 continue # Move to next message in outer loop
             elif str(type(internal_message)) == str(ToolCall):
@@ -163,8 +162,8 @@ def generate(
                     final_text_for_user_message = f"{system_prompt}\n\n{current_message_text}"
                     system_prompt_applied = True # Mark as applied
 
-                message_content_obj = {"type": "text", "text": final_text_for_user_message}
-                openai_message = {"role": role, "content": [message_content_obj]}
+                # For regular text messages, OpenAI expects content as a string
+                openai_message = {"role": role, "content": final_text_for_user_message}
                 openai_messages.append(openai_message)
 
         # If cot_model is True and system_prompt was provided but not applied (e.g., no user messages found, though unlikely for an agent)
@@ -173,7 +172,7 @@ def generate(
             # Or, one might argue it's an error condition for COT if no user prompt exists.
             # For now, let's log a warning and add it as a user message, as some COT models might expect user turn for instructions.
             logger.warning("COT mode: System prompt provided but no initial user message to prepend to. Adding as a separate user message.")
-            openai_messages.insert(0, {"role": "user", "content": [{"type": "text", "text": system_prompt}]})
+            openai_messages.insert(0, {"role": "user", "content": system_prompt})
 
         # Turn tool_choice into OpenAI tool_choice format
         if tool_choice is None:
@@ -234,8 +233,7 @@ def generate(
                     raise e
                 else:
                     print(f"Retrying OpenAI request: {retry + 1}/{self.max_retries}")
-                    # Sleep 8-12 seconds with jitter to avoid thundering herd.
-                    time.sleep(10 * random.uniform(0.8, 1.2))
+                    time.sleep(30)
 
         # Convert messages back to internal format
         internal_messages = []
@@ -248,10 +246,10 @@ def generate(
         content = openai_response_message.content
 
         # Exactly one of tool_calls or content should be present
-        if tool_calls and content:
-            raise ValueError("Only one of tool_calls or content should be present")
-        elif not tool_calls and not content:
-            raise ValueError("Either tool_calls or content should be present")
+        #if tool_calls and content:
+        #    raise ValueError("Only one of tool_calls or content should be present")
+        #elif not tool_calls and not content:
+        #    raise ValueError("Either tool_calls or content should be present")
 
         if tool_calls:
             available_tool_names = {t.name for t in tools} # Get set of known tool names
@@ -297,10 +295,13 @@ def generate(
             if not processed_tool_call:
                 logger.warning("No valid and available tool calls found after filtering.")
 
-        elif content:
-            internal_messages.append(TextResult(text=content))
+        elif content is not None:
+            # Handle empty content (including empty strings) from models that may return reasoning without content
+            internal_messages.append(TextResult(text=content if content else ""))
         else:
-            raise ValueError(f"Unknown message type: {openai_response_message}")
+            # Only raise error if we truly have no content and no tool calls
+            logger.warning(f"Response has no content or tool_calls: {openai_response_message}")
+            internal_messages.append(TextResult(text=""))
 
         assert response.usage is not None
         message_metadata = {

src/ii_agent/migrations/env.py

@@ -81,4 +81,5 @@ def run_migrations_online() -> None:
 if context.is_offline_mode():
     run_migrations_offline()
 else:
-    run_migrations_online()
+    pass 
+    #run_migrations_online()

src/ii_agent/prompts/system_prompt.py

@@ -157,12 +157,9 @@ def get_system_prompt(workspace_mode: WorkSpaceMode):
 </planner_module>
 
 <todo_rules>
-- Create todo.md file as checklist based on task planning from planner module
-- Task planning takes precedence over todo.md, while todo.md contains more details
-- Update markers in todo.md via text replacement tool immediately after completing each item
-- Rebuild todo.md when task planning changes significantly
-- Must use todo.md to record and update progress for information gathering tasks
-- When all planned steps are complete, verify todo.md completion and remove skipped items
+- Use TodoRead and TodoWrite tools to manage your todo list
+- Frequently update the todo list to track the progress
+- When all planned steps are complete, verify todo list completion and remove skipped items
 </todo_rules>
 
 <message_rules>
@@ -351,7 +348,7 @@ def get_system_prompt(workspace_mode: WorkSpaceMode):
 - Events may originate from other system modules; only use explicitly provided tools
 </tool_use_rules>
 
-Today is {datetime.now().strftime("%Y-%m-%d")}. The first step of a task is to use `message_user` tool to plan details of the task. Then regularly update the todo.md file to track the progress.
+Today is {datetime.now().strftime("%Y-%m-%d")}. The first step of a task is to use `message_user` tool to plan details of the task. Then regularly update the todo list to track the progress.
 """