Improvement : Centralised prompt management

.gitignore

@@ -8,7 +8,6 @@ cli/dist
 cli/.venv
 venv/
 .env
-*.json
 .momentum
 db
 cli/momentum_cli/.momentum
@@ -22,3 +21,4 @@ cli/momentum_cli/.momentum
 projects/
 # Ignore PyCharm config
 .idea
+service-account.json

app/alembic/versions/20241127095409_625f792419e7_support_for_repo_path.py

@@ -8,7 +8,6 @@
 
 from typing import Sequence, Union
 
-import sqlalchemy as sa
 from alembic import op
 
 # revision identifiers, used by Alembic.
@@ -19,7 +18,8 @@
 
 
 def upgrade() -> None:
-    op.execute('ALTER TABLE projects ADD COLUMN repo_path TEXT DEFAULT NULL')
+    op.execute("ALTER TABLE projects ADD COLUMN repo_path TEXT DEFAULT NULL")
+
 
 def downgrade() -> None:
     op.drop_column("projects", "repo_path")

app/main.py

@@ -16,9 +16,11 @@
     router as conversations_router,
 )
 from app.modules.intelligence.agents.agents_router import router as agent_router
+from app.modules.intelligence.llm_provider.llm_provider_router import (
+    router as llm_provider_router,
+)
 from app.modules.intelligence.prompts.prompt_router import router as prompt_router
 from app.modules.intelligence.prompts.system_prompt_setup import SystemPromptSetup
-from app.modules.intelligence.provider.provider_router import router as provider_router
 from app.modules.intelligence.tools.tool_router import router as tool_router
 from app.modules.key_management.secret_manager import router as secret_manager_router
 from app.modules.parsing.graph_construction.parsing_router import (
@@ -110,7 +112,9 @@ def include_routers(self):
         self.app.include_router(search_router, prefix="/api/v1", tags=["Search"])
         self.app.include_router(github_router, prefix="/api/v1", tags=["Github"])
         self.app.include_router(agent_router, prefix="/api/v1", tags=["Agents"])
-        self.app.include_router(provider_router, prefix="/api/v1", tags=["Providers"])
+        self.app.include_router(
+            llm_provider_router, prefix="/api/v1", tags=["Providers"]
+        )
         self.app.include_router(tool_router, prefix="/api/v1", tags=["Tools"])
         if os.getenv("isDevelopmentMode") != "enabled":
             self.app.include_router(

app/modules/code_provider/github/github_router.py

@@ -16,19 +16,18 @@ async def get_user_repos(
 ):
     user_repo_list = await GithubController(db).get_user_repos(user=user)
     user_repo_list["repositories"].extend(config_provider.get_demo_repo_list())
-    
+
     # Remove duplicates while preserving order
     seen = set()
     deduped_repos = []
     for repo in reversed(user_repo_list["repositories"]):
         # Create tuple of values to use as hash key
         repo_key = repo["full_name"]
 
-
         if repo_key not in seen:
             seen.add(repo_key)
             deduped_repos.append(repo)
-            
+
     user_repo_list["repositories"] = deduped_repos
     return user_repo_list
 

app/modules/conversations/conversation/conversation_service.py

@@ -34,11 +34,11 @@
 from app.modules.intelligence.agents.custom_agents.custom_agents_service import (
     CustomAgentsService,
 )
-from app.modules.intelligence.memory.chat_history_service import ChatHistoryService
-from app.modules.intelligence.provider.provider_service import (
-    AgentType,
-    ProviderService,
+from app.modules.intelligence.llm_provider.llm_provider_service import (
+    LLMProviderService,
 )
+from app.modules.intelligence.memory.chat_history_service import ChatHistoryService
+from app.modules.intelligence.prompts_provider.agent_types import AgentLLMType
 from app.modules.projects.projects_service import ProjectService
 from app.modules.users.user_service import UserService
 from app.modules.utils.posthog_helper import PostHogClient
@@ -74,7 +74,7 @@ def __init__(
         user_email: str,
         project_service: ProjectService,
         history_manager: ChatHistoryService,
-        provider_service: ProviderService,
+        llm_provider_service: LLMProviderService,
         agent_injector_service: AgentInjectorService,
         custom_agent_service: CustomAgentsService,
     ):
@@ -83,24 +83,24 @@ def __init__(
         self.user_email = user_email
         self.project_service = project_service
         self.history_manager = history_manager
-        self.provider_service = provider_service
+        self.llm_provider_service = llm_provider_service
         self.agent_injector_service = agent_injector_service
         self.custom_agent_service = custom_agent_service
 
     @classmethod
     def create(cls, db: Session, user_id: str, user_email: str):
         project_service = ProjectService(db)
         history_manager = ChatHistoryService(db)
-        provider_service = ProviderService(db, user_id)
-        agent_injector_service = AgentInjectorService(db, provider_service, user_id)
+        llm_provider_service = LLMProviderService(db, user_id)
+        agent_injector_service = AgentInjectorService(db, llm_provider_service, user_id)
         custom_agent_service = CustomAgentsService()
         return cls(
             db,
             user_id,
             user_email,
             project_service,
             history_manager,
-            provider_service,
+            llm_provider_service,
             agent_injector_service,
             custom_agent_service,
         )
@@ -206,7 +206,7 @@ def _create_conversation_record(
         logger.info(
             f"Project id : {conversation.project_ids[0]} Created new conversation with ID: {conversation_id}, title: {title}, user_id: {user_id}, agent_id: {conversation.agent_ids[0]}"
         )
-        provider_name = self.provider_service.get_llm_provider_name()
+        provider_name = self.llm_provider_service.get_llm_provider_name()
         PostHogClient().send_event(
             user_id,
             "create Conversation Event",
@@ -261,7 +261,7 @@ async def store_message(
                 conversation_id, message_type, user_id
             )
             logger.info(f"Stored message in conversation {conversation_id}")
-            provider_name = self.provider_service.get_llm_provider_name()
+            provider_name = self.llm_provider_service.get_llm_provider_name()
 
             PostHogClient().send_event(
                 user_id,
@@ -335,7 +335,7 @@ async def _generate_title(
     ) -> str:
         agent_type = conversation.agent_ids[0]
 
-        llm = self.provider_service.get_small_llm(agent_type=AgentType.LANGCHAIN)
+        llm = self.llm_provider_service.get_small_llm(agent_type=AgentLLMType.LANGCHAIN)
         prompt = ChatPromptTemplate.from_template(
             "Given an agent type '{agent_type}' and an initial message '{message}', "
             "generate a concise and relevant title for a conversation. "

app/modules/intelligence/agents/agent_injector_service.py

@@ -24,26 +24,28 @@
 from app.modules.intelligence.agents.custom_agents.custom_agents_service import (
     CustomAgentsService,
 )
-from app.modules.intelligence.provider.provider_service import (
-    AgentType,
-    ProviderService,
+from app.modules.intelligence.llm_provider.llm_provider_service import (
+    LLMProviderService,
 )
+from app.modules.intelligence.prompts_provider.agent_types import AgentLLMType
 
 logger = logging.getLogger(__name__)
 
 
 class AgentInjectorService:
-    def __init__(self, db: Session, provider_service: ProviderService, user_id: str):
+    def __init__(self, db: Session, provider_service: LLMProviderService, user_id: str):
         self.sql_db = db
         self.provider_service = provider_service
         self.custom_agent_service = CustomAgentsService()
         self.agents = self._initialize_agents()
         self.user_id = user_id
 
     def _initialize_agents(self) -> Dict[str, Any]:
-        mini_llm = self.provider_service.get_small_llm(agent_type=AgentType.LANGCHAIN)
+        mini_llm = self.provider_service.get_small_llm(
+            agent_type=AgentLLMType.LANGCHAIN
+        )
         reasoning_llm = self.provider_service.get_large_llm(
-            agent_type=AgentType.LANGCHAIN
+            agent_type=AgentLLMType.LANGCHAIN
         )
         return {
             "debugging_agent": DebuggingChatAgent(mini_llm, reasoning_llm, self.sql_db),
@@ -66,7 +68,7 @@ def get_agent(self, agent_id: str) -> Any:
             return self.agents[agent_id]
         else:
             reasoning_llm = self.provider_service.get_large_llm(
-                agent_type=AgentType.LANGCHAIN
+                agent_type=AgentLLMType.LANGCHAIN
             )
             return CustomAgent(
                 llm=reasoning_llm,

app/modules/intelligence/agents/agents/blast_radius_agent.py

@@ -5,10 +5,11 @@
 from pydantic import BaseModel, Field
 
 from app.modules.conversations.message.message_schema import NodeContext
-from app.modules.intelligence.provider.provider_service import (
-    AgentType,
-    ProviderService,
+from app.modules.intelligence.llm_provider.llm_provider_service import (
+    LLMProviderService,
 )
+from app.modules.intelligence.prompts_provider.agent_prompts import AgentPromptsProvider
+from app.modules.intelligence.prompts_provider.agent_types import AgentLLMType
 from app.modules.intelligence.tools.change_detection.change_detection_tool import (
     ChangeDetectionResponse,
     get_change_detection_tool,
@@ -33,10 +34,13 @@ def __init__(self, sql_db, user_id, llm):
         )
 
     async def create_agents(self):
+        agent_prompt = AgentPromptsProvider.get_agent_prompt(
+            agent_id="blast_radius_agent", agent_type=AgentLLMType.CREWAI
+        )
         blast_radius_agent = Agent(
-            role="Blast Radius Agent",
-            goal="Explain the blast radius of the changes made in the code.",
-            backstory="You are an expert in understanding the impact of code changes on the codebase.",
+            role=agent_prompt["role"],
+            goal=agent_prompt["goal"],
+            backstory=agent_prompt["backstory"],
             allow_delegation=False,
             verbose=True,
             llm=self.llm,
@@ -60,50 +64,17 @@ async def create_tasks(
         query: str,
         blast_radius_agent,
     ):
+        task_prompt = AgentPromptsProvider.get_task_prompt(
+            task_id="analyze_changes_task",
+            agent_type=AgentLLMType.CREWAI,
+            project_id=project_id,
+            query=query,
+            ChangeDetectionResponse=ChangeDetectionResponse,
+            BlastRadiusAgentResponse=self.BlastRadiusAgentResponse,
+        )
+
         analyze_changes_task = Task(
-            description=f"""Fetch the changes in the current branch for project {project_id} using the get code changes tool.
-            The response of the fetch changes tool is in the following format:
-            {ChangeDetectionResponse.model_json_schema()}
-            In the response, the patches contain the file patches for the changes.
-            The changes contain the list of changes with the updated and entry point code. Entry point corresponds to the API/Consumer upstream of the function that the change was made in.
-            The citations contain the list of file names referenced in the changed code and entry point code.
-
-            You also have access the the query knowledge graph tool to answer natural language questions about the codebase during the analysis.
-            Based on the response from the get code changes tool, formulate queries to ask details about specific changed code elements.
-            1. Frame your query for the knowledge graph tool:
-            - Identify key concepts, code elements, and implied relationships from the changed code.
-            - Consider the context from the users query: {query}.
-            - Determine the intent and key technical terms.
-            - Transform into keyword phrases that might match docstrings:
-                * Use concise, functionality-based phrases (e.g., "creates document MongoDB collection").
-                * Focus on verb-based keywords (e.g., "create", "define", "calculate").
-                * Include docstring-related keywords like "parameters", "returns", "raises" when relevant.
-                * Preserve key technical terms from the original query.
-                * Generate multiple keyword variations to increase matching chances.
-                * Be specific in keywords to improve match accuracy.
-                * Ensure the query includes relevant details and follows a similar structure to enhance similarity search results.
-
-            2. Execute your formulated query using the knowledge graph tool.
-
-            Analyze the changes fetched and explain their impact on the codebase. Consider the following:
-            1. Which functions or classes have been directly modified?
-            2. What are the potential side effects of these changes?
-            3. Are there any dependencies that might be affected?
-            4. How might these changes impact the overall system behavior?
-            5. Based on the entry point code, determine which APIs or consumers etc are impacted by the changes.
-
-            Refer to the {query} for any specific instructions and follow them.
-
-            Based on the analysis, provide a structured inference of the blast radius:
-            1. Summarize the direct changes
-            2. List potential indirect effects
-            3. Identify any critical areas that require careful testing
-            4. Suggest any necessary refactoring or additional changes to mitigate risks
-            6. If the changes are impacting multiple APIs/Consumers, then say so.
-
-
-            Ensure that your output ALWAYS follows the structure outlined in the following pydantic model:
-            {self.BlastRadiusAgentResponse.model_json_schema()}""",
+            description=task_prompt,
             expected_output=f"Comprehensive impact analysis of the code changes on the codebase and answers to the users query about them. Ensure that your output ALWAYS follows the structure outlined in the following pydantic model : {self.BlastRadiusAgentResponse.model_json_schema()}",
             agent=blast_radius_agent,
             tools=[
@@ -140,8 +111,8 @@ async def run(
 async def kickoff_blast_radius_agent(
     query: str, project_id: str, node_ids: List[NodeContext], sql_db, user_id, llm
 ) -> Dict[str, str]:
-    provider_service = ProviderService(sql_db, user_id)
-    crew_ai_mini_llm = provider_service.get_small_llm(agent_type=AgentType.CREWAI)
+    provider_service = LLMProviderService(sql_db, user_id)
+    crew_ai_mini_llm = provider_service.get_small_llm(agent_type=AgentLLMType.CREWAI)
     blast_radius_agent = BlastRadiusAgent(sql_db, user_id, crew_ai_mini_llm)
     result = await blast_radius_agent.run(project_id, node_ids, query)
     return result