feat: add support for aider

Aider offers two types of integrations:
- compatible OPENAI API
- local ollama

Add support for those 2 integrations, and correct all different nuances
that we are getting with this new engine

Closes: #440

.gitignore

@@ -49,3 +49,4 @@ sqlite_data/vectordb.db
 
 # certificate directory
 *certs/
+.aider*

src/codegate/pipeline/base.py

@@ -253,9 +253,11 @@ def get_latest_user_messages(request: ChatCompletionRequest) -> str:
 
         for message in reversed(request.get("messages", [])):
             if message["role"] == "user":
-                latest_user_messages += "\n" + message["content"]
-            else:
-                break
+                # if found we can stop here, if not we continue until we find it
+                message_str = message.get("content", "")
+                if message_str:
+                    latest_user_messages += "\n" + str(message_str)
+                    break
 
         return latest_user_messages
 

src/codegate/pipeline/codegate_context_retriever/codegate.py

@@ -59,7 +59,6 @@ async def process(
         """
         Use RAG DB to add context to the user request
         """
-
         # Get the latest user messages
         user_messages = self.get_latest_user_messages(request)
 

src/codegate/pipeline/output.py

@@ -27,6 +27,8 @@ class OutputPipelineContext:
     snippets: List[CodeSnippet] = field(default_factory=list)
     # Store all content that has been processed by the pipeline
     processed_content: List[str] = field(default_factory=list)
+    # partial buffer to store prefixes
+    prefix_buffer: str = ""
 
 
 class OutputPipelineStep(ABC):

src/codegate/pipeline/secrets/secrets.py

@@ -280,7 +280,7 @@ async def process(
             if "content" in message and message["content"]:
                 # Protect the text
                 protected_string, redacted_count = self._redact_text(
-                    message["content"], secrets_manager, session_id, context
+                    str(message["content"]), secrets_manager, session_id, context
                 )
                 new_request["messages"][i]["content"] = protected_string
 
@@ -389,12 +389,17 @@ async def process_chunk(
             return [chunk]
 
         # If we have a partial marker at the end, keep buffering
-        if self.marker_start in buffered_content or self._is_partial_marker_prefix(
-            buffered_content
-        ):
+        if self.marker_start in buffered_content:
+            context.prefix_buffer = ""
+            return []
+
+        if self._is_partial_marker_prefix(buffered_content):
+            context.prefix_buffer += buffered_content
             return []
 
         # No markers or partial markers, let pipeline handle the chunk normally
+        chunk.choices[0].delta.content = context.prefix_buffer + chunk.choices[0].delta.content
+        context.prefix_buffer = ""
         return [chunk]