Fix/biobuddy (#88)

compose.yaml

@@ -132,7 +132,7 @@ services:
       dockerfile: build/Dockerfile
     command: --host=0.0.0.0 --port=5739
   biobuddy:
-    image: 'ghcr.io/basedlabs/biobuddy:1.2.0'
+    image: 'ghcr.io/basedlabs/biobuddy:1.2.1'
     network_mode: host
     build:
       context: microservices/biobuddy
@@ -167,7 +167,7 @@ services:
       dockerfile: build/Dockerfile
     command: --host=0.0.0.0 --port=5740 --workers=1
   nolabs:
-    image: 'ghcr.io/basedlabs/nolabs:2.1.6'
+    image: 'ghcr.io/basedlabs/nolabs:2.1.7'
     network_mode: host
     build:
       context: .

frontend/src/features/biobuddy/BioBuddyChat.vue

@@ -129,6 +129,13 @@ export default defineComponent({
             });
           }
 
+          const researchMatches = this.currentMessageBuffer.match(/<RESEARCH/g);
+          if (researchMatches) {
+            researchMatches.forEach((match) => {
+              this.currentMessageBuffer = this.currentMessageBuffer.replace(/<RESEARCH/g, '');
+            });
+          }
+
           // Process WORKFLOW tags
           const workflowMatches = this.currentMessageBuffer.match(/<WORKFLOW>([\s\S]*?)<END_WORKFLOW>/g);
           if (workflowMatches) {
@@ -537,4 +544,4 @@ export default defineComponent({
   vertical-align: top;
   /* Aligns icon with text if necessary */
 }
-</style>
+</style>

microservices/biobuddy/biobuddy/api.py

@@ -3,7 +3,7 @@
 from biobuddy.services import send_message, send_message_async, invoke_action
 from biobuddy.api_models import SendMessageToBioBuddyRequest, SendMessageToBioBuddyResponse, IsJobRunningResponse, \
     SendActionCallRequest
-from typing import Dict, List, Any
+from typing import Dict
 import json
 
 app = FastAPI(
@@ -15,7 +15,6 @@
 
 connection_manager = ConnectionManager()
 
-from biobuddy.loggers import Log
 from biobuddy.job_state_manager import job_state_manager
 
 @app.post("/send-message")
@@ -44,6 +43,7 @@ async def invoke_function(
 async def websocket_endpoint(websocket: WebSocket):
     await connection_manager.connect(websocket)
     try:
+        print("Original Websocket: ", websocket)
         while True:
             data = await websocket.receive_text()
             request_data = json.loads(data)
@@ -55,7 +55,7 @@ async def websocket_endpoint(websocket: WebSocket):
             else:
                 request = SendMessageToBioBuddyRequest(**request_data)
                 stop_tokens[request.experiment_id] = False
-                async for message in send_message_async(request, stop_tokens):
+                async for message in send_message_async(request, stop_tokens, websocket):
                     if stop_tokens.get(request.experiment_id):
                         await websocket.send_text(json.dumps({"reply_type": "final", "content": "<STOP>"}))
                         break

microservices/biobuddy/biobuddy/gpt_researcher/context/compression.py

@@ -31,9 +31,9 @@ def _get_contextual_retriever(self):
         return contextual_retriever
 
     def _pretty_print_docs(self, docs, top_n):
-        return f"\n".join(f"Source: {d.metadata.get('source')}\n"
-                          f"Title: {d.metadata.get('title')}\n"
-                          f"Content: {d.page_content}\n"
+        return f" \n".join(f"Source: {d.metadata.get('source')}\n "
+                          f"Title: {d.metadata.get('title')}\n "
+                          f"Content: {d.page_content}\n "
                           for i, d in enumerate(docs) if i < top_n)
 
     def get_context(self, query, max_results=5):

microservices/biobuddy/biobuddy/gpt_researcher/master/agent.py

@@ -40,6 +40,7 @@ async def run(self):
         print(f"🔎 Running research for '{self.query}'...")
         # Generate Agent
         self.agent, self.role = await choose_agent(self.query, self.cfg)
+        await asyncio.sleep(0)
         await stream_output("logs", self.agent, self.websocket)
 
         # If specified, the researcher will use the given urls as the context for the research.
@@ -51,7 +52,7 @@ async def run(self):
         # Write Research Report
         if self.report_type == "custom_report":
             self.role = self.cfg.agent_role if self.cfg.agent_role else self.role
-        await stream_output("logs", f"✍️ Writing {self.report_type} for research task: {self.query}...", self.websocket)
+        await stream_output("logs", f"✍️ Writing {self.report_type} for research task: {self.query}...\n", self.websocket)
         report = await generate_report(query=self.query, context=self.context,
                                        agent_role_prompt=self.role, report_type=self.report_type,
                                        websocket=self.websocket, cfg=self.cfg)
@@ -64,7 +65,7 @@ async def get_context_by_urls(self, urls):
         """
         new_search_urls = await self.get_new_urls(urls)
         await stream_output("logs",
-                            f"🧠 I will conduct my research based on the following urls: {new_search_urls}...",
+                            f"\n 🧠 I will conduct my research based on the following urls: {new_search_urls}... \n",
                             self.websocket)
         scraped_sites = scrape_urls(new_search_urls, self.cfg)
         return await self.get_similar_content_by_query(self.query, scraped_sites)
@@ -79,15 +80,15 @@ async def get_context_by_search(self, query):
         # Generate Sub-Queries including original query
         sub_queries = await get_sub_queries(query, self.role, self.cfg) + [query]
         await stream_output("logs",
-                            f"🧠 I will conduct my research based on the following queries: {sub_queries}...",
+                            f"\n 🧠 I will conduct my research based on the following queries: {sub_queries}... \n",
                             self.websocket)
 
         # Run Sub-Queries
         for sub_query in sub_queries:
-            await stream_output("logs", f"\n🔎 Running research for '{sub_query}'...", self.websocket)
+            await stream_output("logs", f"\n 🔎 Running research for '{sub_query}'... \n", self.websocket)
             scraped_sites = await self.scrape_sites_by_query(sub_query)
             content = await self.get_similar_content_by_query(sub_query, scraped_sites)
-            await stream_output("logs", f"📃 {content}", self.websocket)
+            await stream_output("logs", f"\n 📃 {content} \n", self.websocket)
             context.append(content)
 
         return context
@@ -101,7 +102,7 @@ async def get_new_urls(self, url_set_input):
         new_urls = []
         for url in url_set_input:
             if url not in self.visited_urls:
-                await stream_output("logs", f"✅ Adding source url to research: {url}\n", self.websocket)
+                await stream_output("logs", f"\n ✅ Adding source url to research: [{url}]({url}) \n", self.websocket)
 
                 self.visited_urls.add(url)
                 new_urls.append(url)
@@ -129,12 +130,12 @@ async def scrape_sites_by_query(self, sub_query):
 
         # Scrape Urls
         # await stream_output("logs", f"📝Scraping urls {new_search_urls}...\n", self.websocket)
-        await stream_output("logs", f"🤔Researching for relevant information...\n", self.websocket)
+        await stream_output("logs", f"\n 🤔Researching for relevant information... \n", self.websocket)
         scraped_content_results = scrape_urls(new_search_urls, self.cfg)
         return scraped_content_results
 
     async def get_similar_content_by_query(self, query, pages):
-        await stream_output("logs", f"📃 Getting relevant content based on query: {query}...", self.websocket)
+        await stream_output("logs", f"\n 📃 Getting relevant content based on query: {query}... \n", self.websocket)
         # Summarize Raw Data
         context_compressor = ContextCompressor(documents=pages, embeddings=self.memory.get_embeddings())
         # Run Tasks

microservices/biobuddy/biobuddy/gpt_researcher/master/functions.py

@@ -4,6 +4,8 @@
 from .prompts import *
 import json
 
+from ...api_models import SendMessageToBioBuddyResponse
+
 
 def get_retriever(retriever):
     """
@@ -242,7 +244,10 @@ async def stream_output(type, output, websocket=None, logging=True):
         None
     """
     if not websocket or logging:
-        print(output)
+        print("Static output: ", output)
 
     if websocket:
-        await websocket.send_json({"type": type, "output": output})
+        await asyncio.sleep(0)
+        message = SendMessageToBioBuddyResponse(reply_type="stream", content=output)
+        await websocket.send_text(json.dumps(message.to_dict()))
+        await asyncio.sleep(0)

microservices/biobuddy/biobuddy/gpt_researcher/run.py

@@ -1,7 +1,9 @@
+from websocket import WebSocket
+
 from .master.agent import GPTResearcher
 
 
-async def get_report(query: str, report_type: str = 'research_report') -> str:
+async def get_report(query: str, websocket: WebSocket, report_type: str = 'research_report') -> str:
     """
     Get the report for the given query and report type:
     Args:
@@ -10,6 +12,7 @@ async def get_report(query: str, report_type: str = 'research_report') -> str:
     Output:
         report: The generated report in markdown format
     """
-    researcher = GPTResearcher(query, report_type, config_path="../config.json")
+    researcher = GPTResearcher(query, report_type, config_path="../config.json", websocket=websocket)
+    print("GPT researcher")
     report = await researcher.run()
     return report

microservices/biobuddy/biobuddy/gpt_researcher/utils/llm.py

@@ -7,6 +7,7 @@
 from typing import Optional
 
 from ..master.prompts import auto_agent_instructions
+from ...api_models import SendMessageToBioBuddyResponse
 
 
 async def create_chat_completion(
@@ -85,7 +86,8 @@ async def stream_response(model, messages, temperature, max_tokens, llm_provider
             paragraph += content
             if "\n" in paragraph:
                 if websocket is not None:
-                    await websocket.send_json({"type": "report", "output": paragraph})
+                    message = SendMessageToBioBuddyResponse(reply_type="stream", content=paragraph)
+                    await websocket.send_text(json.dumps(message.to_dict()))
                 else:
                     print(f"{Fore.GREEN}{paragraph}{Style.RESET_ALL}")
                 paragraph = ""

microservices/biobuddy/biobuddy/prompts.py

@@ -9,8 +9,11 @@ def generate_system_prompt() -> str:
 
 
 def generate_strategy_prompt(tools_description: str, query: str, available_components: str, current_workflow: str) -> str:
-    return (f"Given the available tools ({tools_description}), decide whether a direct reply is sufficient "
-            f"or if a specific plan involving these tools is necessary. "
+    return (f"Given the available tools ({tools_description}), decide whether a direct reply or a research is sufficient "
+            f"or if a specific plan involving these tools is necessary."
+            f"If a user asks to do some online literature research directly, such as 'Could you find me some information about latest research on GFP' or 'What are the latest news abuot rhodopsins?' or 'Can you search me some information about vaccines' etc., "
+            f"then reply just with one word-tag '<RESEARCH>'. Do it only if they ask you to search the information, search the Internet or some up-to-date info."
+            f"Otherwise, follow these instructions:"
             f"If calling function calls are not needed, provide the direct reply (without stating that it's a direct reply, just the text of the reply). "
             f"If the query explicitly asks to pull specific data (such as named proteins or ligands), reply with a plan of all actions required based on the query and tools descriptions in the format (have as many actions as needed): "
             f"\"<ACTION> 1. Call tool_1 in order to do X <END_ACTION> <ACTION> 2. Call tool_2 to achieve Y <END_ACTION> <ACTION> 3. Call tool_3 to do something else <END_ACTION>\". "

microservices/biobuddy/biobuddy/services.py

@@ -2,6 +2,7 @@
 
 from langchain.callbacks import AsyncIteratorCallbackHandler
 from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
+from websocket import WebSocket
 
 from biobuddy.api_models import SendMessageToBioBuddyResponse, SendMessageToBioBuddyRequest
 from biobuddy.prompts import generate_strategy_prompt, generate_system_prompt, generate_workflow_prompt
@@ -83,7 +84,10 @@ def send_message(request: SendMessageToBioBuddyRequest) -> SendMessageToBioBuddy
     )
 
 
-async def send_message_async(request: SendMessageToBioBuddyRequest, stop_tokens: Dict[str, bool]) -> AsyncGenerator[SendMessageToBioBuddyResponse, None]:
+async def send_message_async(request: SendMessageToBioBuddyRequest,
+                             stop_tokens: Dict[str, bool],
+                             websocket: WebSocket
+                             ) -> AsyncGenerator[SendMessageToBioBuddyResponse, None]:
     system_message_content = generate_system_prompt()
     history_messages = [SystemMessage(content=system_message_content)]
     for msg in request.previous_messages:
@@ -92,6 +96,8 @@ async def send_message_async(request: SendMessageToBioBuddyRequest, stop_tokens:
         elif msg['role'] == 'assistant':
             history_messages.append(AIMessage(content=msg['content']))
 
+    # await get_report(request.message_content, websocket)
+
     tools_description = " ".join([f"{(tool['function']['name'], tool['function']['description'])}, " for tool in request.tools])
     strategy_prompt = generate_strategy_prompt(tools_description, request.message_content, str(request.available_components), str(request.current_workflow))
 
@@ -117,7 +123,9 @@ async def chat_with_model():
             break
         response_buffer += response
 
-        print(response_buffer)
+        if "<RESEARCH>" in response_buffer:
+            await get_report(request.message_content, websocket)
+            response_buffer = response_buffer.replace("<RESEARCH>", "")
 
         yield SendMessageToBioBuddyResponse(reply_type="stream", content=response)
 

microservices/biobuddy/requirements.txt

@@ -5,12 +5,12 @@ openapi
 openai
 biopython
 langchain-openai
-langchain
+langchain==0.1.14
 websockets
 asyncio==3.4.3
 beautifulsoup4==4.12.2
 colorama==0.4.6
-duckduckgo_search==5.3.0
+duckduckgo_search
 tavily-python==0.3.1
 arxiv==2.0.0
 PyMuPDF==1.23.6