feat: first interation for kodu (#778)

* feat: first interation for kodu

* fix secrets in kodu

* fix documentation

Author: yrobla

README.md

@@ -91,6 +91,12 @@ With Cline, you can choose between differnet leading AI model providers:
 - 💻 [LM Studio](https://lmstudio.ai/)
 - 💻 Local LLMs with [Ollama](https://ollama.com/)
 
+- **[Kodu](https://kodu.ai)**
+
+You can use Kodu with OpenAI compatible APIs:
+
+- 🧠 [OpenAI API](https://openai.com/api/)
+
 
 ### Privacy first
 

src/codegate/pipeline/cli/cli.py

@@ -82,7 +82,7 @@ async def process(
             base_tool = get_tool_name_from_messages(request)
             codegate_regex = re.compile(r"^codegate(?:\s+(.*))?", re.IGNORECASE)
 
-            if base_tool and base_tool == "cline":
+            if base_tool and base_tool in ["cline", "kodu"]:
                 # Check if there are <task> or <feedback> tags
                 tag_match = re.search(
                     r"<(task|feedback)>(.*?)</\1>", last_user_message_str, re.DOTALL
@@ -111,7 +111,7 @@ async def process(
                 if args:
                     context.shortcut_response = True
                     cmd_out = await codegate_cli(args[1:])
-                    if base_tool and base_tool == "cline":
+                    if base_tool in ["cline", "kodu"]:
                         cmd_out = (
                             f"<attempt_completion><result>{cmd_out}</result></attempt_completion>\n"
                         )

src/codegate/pipeline/codegate_context_retriever/codegate.py

@@ -13,7 +13,7 @@
 from codegate.pipeline.extract_snippets.extract_snippets import extract_snippets
 from codegate.storage.storage_engine import StorageEngine
 from codegate.utils.package_extractor import PackageExtractor
-from codegate.utils.utils import generate_vector_string
+from codegate.utils.utils import generate_vector_string, get_tool_name_from_messages
 
 logger = structlog.get_logger("codegate")
 
@@ -63,7 +63,7 @@ async def process(
         last_message = self.get_last_user_message_block(request)
         if not last_message:
             return PipelineResult(request=request)
-        user_message, _ = last_message
+        user_message, last_user_idx = last_message
 
         # Create storage engine object
         storage_engine = StorageEngine()
@@ -124,33 +124,34 @@ async def process(
             context_str = self.generate_context_str(all_bad_packages, context)
             context.bad_packages_found = True
 
-            last_user_idx = self.get_last_user_message_idx(request)
-
             # Make a copy of the request
             new_request = request.copy()
 
-            # Format: "Context: {context_str} \n Query: {last user message content}"
-            message = new_request["messages"][last_user_idx]
-            message_str = str(message["content"])  # type: ignore
-            # Add the context to the last user message
-            if message_str.strip().startswith("<task>"):
-                # formatting of cline
-                match = re.match(r"(<task>)(.*?)(</task>)(.*)", message_str, re.DOTALL)
-                if match:
-                    task_start, task_content, task_end, rest_of_message = match.groups()
-
-                # Embed the context into the task block
-                updated_task_content = (
-                    f"{task_start}Context: {context_str}\n"
-                    + f"Query: {task_content.strip()}</details>{task_end}"
-                )
-
-                # Combine the updated task block with the rest of the message
-                context_msg = updated_task_content + rest_of_message
-            else:
-                context_msg = f"Context: {context_str} \n\n Query: {message_str}"  # type: ignore
-            message["content"] = context_msg
-
-            logger.debug("Final context message", context_message=context_msg)
-
+            # perform replacement in all the messages starting from this index
+            for i in range(last_user_idx, len(new_request["messages"])):
+                message = new_request["messages"][i]
+                message_str = str(message["content"])  # type: ignore
+                context_msg = message_str
+                # Add the context to the last user message
+                base_tool = get_tool_name_from_messages(request)
+                if base_tool in ["cline", "kodu"]:
+                    match = re.search(r"<task>\s*(.*?)\s*</task>(.*)", message_str, re.DOTALL)
+                    if match:
+                        task_content = match.group(1)  # Content within <task>...</task>
+                        rest_of_message = match.group(2).strip()  # Content after </task>, if any
+
+                        # Embed the context into the task block
+                        updated_task_content = (
+                            f"<task>Context: {context_str}"
+                            + f"Query: {task_content.strip()}</task>"
+                        )
+
+                        # Combine updated task content with the rest of the message
+                        context_msg = updated_task_content + rest_of_message
+
+                else:
+                    context_msg = f"Context: {context_str} \n\n Query: {message_str}"  # type: ignore
+
+                new_request["messages"][i]["content"] = context_msg
+                logger.debug("Final context message", context_message=context_msg)
             return PipelineResult(request=new_request, context=context)

src/codegate/pipeline/secrets/secrets.py

@@ -450,18 +450,22 @@ async def process_chunk(
             or input_context.metadata.get("redacted_secrets_count", 0) == 0
         ):
             return [chunk]
-
-        is_cline_client = any(
-            "Cline" in str(message.trigger_string or "")
-            for message in input_context.alerts_raised or []
+        tool_name = next(
+            (
+                tool.lower()
+                for tool in ["Cline", "Kodu"]
+                for message in input_context.alerts_raised or []
+                if tool in str(message.trigger_string or "")
+            ),
+            "",
         )
 
         # Check if this is the first chunk (delta role will be present, others will not)
         if len(chunk.choices) > 0 and chunk.choices[0].delta.role:
             redacted_count = input_context.metadata["redacted_secrets_count"]
             secret_text = "secret" if redacted_count == 1 else "secrets"
             # Create notification chunk
-            if is_cline_client:
+            if tool_name in ["cline", "kodu"]:
                 notification_chunk = self._create_chunk(
                     chunk,
                     f"<thinking>\n🛡️ [CodeGate prevented {redacted_count} {secret_text}]"

src/codegate/pipeline/system_prompt/codegate.py

@@ -103,4 +103,41 @@ async def process(
             # Update the existing system prompt
             request_system_message["content"] = system_prompt
 
+        # check if we are in kodu
+        if "</kodu_action>" in new_request.get("stop", []):
+            # Collect messages from the assistant matching the criteria
+            relevant_contents = [
+                message["content"]
+                for message in new_request["messages"]
+                if message["role"] == "assistant"
+                and (
+                    message["content"].startswith("**Warning")
+                    or message["content"].startswith("<thinking>")
+                )
+            ]
+
+            if relevant_contents:
+                # Combine the contents into a single message
+                summarized_content = (
+                    "<attempt_completion><result>"
+                    + "".join(relevant_contents)
+                    + "</result></attempt_completion>"
+                )
+
+                # Replace the messages with a single summarized message
+                new_request["messages"] = [
+                    message
+                    for message in new_request["messages"]
+                    if not (
+                        message["role"] == "assistant"
+                        and (
+                            message["content"].startswith("**Warning")
+                            or message["content"].startswith("<thinking>")
+                        )
+                    )
+                ]
+
+                # Append the summarized message to the messages
+                new_request["messages"].append({"role": "assistant", "content": summarized_content})
+
         return PipelineResult(request=new_request, context=context)

src/codegate/providers/base.py

@@ -17,6 +17,7 @@
 from codegate.providers.formatting.input_pipeline import PipelineResponseFormatter
 from codegate.providers.normalizer.base import ModelInputNormalizer, ModelOutputNormalizer
 from codegate.providers.normalizer.completion import CompletionNormalizer
+from codegate.utils.utils import get_tool_name_from_messages
 
 logger = structlog.get_logger("codegate")
 
@@ -233,12 +234,7 @@ async def complete(
         # Execute the completion and translate the response
         # This gives us either a single response or a stream of responses
         # based on the streaming flag
-        is_cline_client = any(
-            "Cline" in str(message.get("content", "")) for message in data.get("messages", [])
-        )
-        base_tool = ""
-        if is_cline_client:
-            base_tool = "cline"
+        base_tool = get_tool_name_from_messages(data)
 
         model_response = await self._completion_handler.execute_completion(
             provider_request,

src/codegate/utils/utils.py

@@ -45,9 +45,7 @@ def get_tool_name_from_messages(data):
     Returns:
         str: The name of the tool found in the messages, or None if no match is found.
     """
-    tools = [
-        "Cline",
-    ]
+    tools = ["Cline", "Kodu"]
     for message in data.get("messages", []):
         message_content = str(message.get("content", ""))
         for tool in tools: