Upload folder using huggingface_hub

app.py

@@ -43,18 +43,24 @@ logger = logging.getLogger("agent-app")
 PLANNING_SYSTEM_PROMPT = """You are an autonomous planning agent assisting in project ideation.
 Your goal is to help the user refine their idea into a concrete project structure.
 
-### Sequential Thinking
-You MUST use the `sequentialthinking` tool to:
-1. Break down the user's initial idea into logical components.
+### Predefined Pipeline: Sequential Thinking
+You MUST start your response by using the `sequentialthinking` tool to:
+1. Analyze the user's initial idea and break it down into logical components.
 2. Reflect on the feasibility and security of the architecture.
 3. Reason about potential API integrations and future use cases.
 4. Iterate on your thoughts until you have a complete vision.
 
 ### Process
-- Start by using `sequentialthinking` to analyze the project idea.
-- Continue thinking until you have covered all aspects of the required structured fields (Description, Tasks, Repos, etc.).
-- Once you are finished thinking, provide your final response to the user.
-- The system will automatically extract the structured data from your final response.
+- You MUST use `sequentialthinking` at the beginning of the session based on the idea given.
+- Continue thinking until you have covered all aspects of the required structured fields:
+  1. Project Description (vision, goals, integrations)
+  2. Tasks and Tests (following the 7-point template)
+  3. GitHub Repos
+  4. Functionality Expectations
+  5. API Endpoints
+  6. Huggingface Deployment Data
+- Once you are finished thinking, provide your final answer to the user.
+- After your final answer, the session will automatically close and populate the structured fields.
 
 ### Guidelines
 - Be thorough, professional, and visionary.
@@ -117,7 +123,7 @@ HELMHOLTZ_BASE_URL = "https://api.helmholtz-blablador.fz-juelich.de/v1"
 api_key = os.environ.get("BLABLADOR_API_KEY") or "EMPTY_KEY"
 
 chat_llm = ChatOpenAI(
-    model="alias-large",
+    model="alias-fast",
     base_url=HELMHOLTZ_BASE_URL,
     api_key=api_key,
     max_tokens=2048,
@@ -125,7 +131,7 @@ chat_llm = ChatOpenAI(
 )
 
 code_llm = ChatOpenAI(
-    model="alias-code",
+    model="alias-fast",
     base_url=HELMHOLTZ_BASE_URL,
     api_key=api_key,
     max_tokens=1024,
@@ -141,13 +147,21 @@ fast_llm = ChatOpenAI(
 )
 
 huge_llm = ChatOpenAI(
-    model="alias-huge",
+    model="alias-fast",
     base_url=HELMHOLTZ_BASE_URL,
     api_key=api_key,
     max_tokens=1024,
     max_retries=0
 )
 
+large_llm = ChatOpenAI(
+    model="alias-large",
+    base_url=HELMHOLTZ_BASE_URL,
+    api_key=api_key,
+    max_tokens=2048,
+    max_retries=0
+)
+
 async def ainvoke_with_retry(llm, prompt, max_retries=5, delay=30):
     """Wrapper for LLM calls with retry logic to handle intermittent provider errors."""
     last_error = None
@@ -786,11 +800,13 @@ async def handle_close_ideate(history, persona="planning"):
     }
 
     extracted = {}
+    extraction_llm = large_llm if persona == "mentor" else fast_llm
+    
     for key, instruction in fields_config.items():
         try:
             prompt = f"{instruction}\n\nLOG:\n{chat_text}\n\nRespond ONLY with the extracted content for this field. Do not add conversational filler."
-            logger.info(f"Extracting {key} using alias-large...")
-            response = await ainvoke_with_retry(chat_llm, prompt)
+            logger.info(f"Extracting {key} for {persona}...")
+            response = await ainvoke_with_retry(extraction_llm, prompt)
             content = response.content
             if "<think>" in content:
                 content = content.split("</think>")[-1].strip()
@@ -1023,6 +1039,10 @@ async def handle_supervisor_nudge(session_id, log_file=""):
     if not session_id:
         return "No session ID."
 
+    # Wait 180 seconds as requested before retrieving activities
+    logger.info(f"Supervisor waiting 180s before retrieving activities for {session_id}...")
+    await asyncio.sleep(180)
+
     tools = await get_all_tools()
 
     # Load context from log_file if provided
@@ -1063,7 +1083,7 @@ async def handle_supervisor_nudge(session_id, log_file=""):
         "Respond with the EXACT message you want to send to Jules."
     )
 
-    response = await ainvoke_with_retry(chat_llm, [
+    response = await ainvoke_with_retry(fast_llm, [
         {"role": "system", "content": system_msg},
         {"role": "user", "content": prompt}
     ])
@@ -1131,8 +1151,8 @@ async def load_github_branches(repo_url, default_owner="JsonLord"):
         logger.error(f"Error loading branches: {e}")
         return gr.update(choices=["main"], value="main")
 
-async def handle_jules_comm(repo_url, branch="main", log_file="", hf_prof_fallback="", hf_space_fallback="", hf_token_fallback=""):
-    if not repo_url:
+async def handle_jules_comm(repo_url_input, branch="main", log_file="", hf_prof_fallback="", hf_space_fallback="", hf_token_fallback=""):
+    if not repo_url_input:
         return "No repository URL/source provided.", ""
 
     all_tools = await get_all_tools()
@@ -1144,7 +1164,13 @@ async def handle_jules_comm(repo_url, branch="main", log_file="", hf_prof_fallba
     ]
     tools = [t for t in all_tools if t.name in jules_tool_names]
     
-    repo_url = transform_github_url(repo_url)
+    # Ensure we have the full HTTPS URL for instructions
+    full_repo_url = repo_url_input
+    if not full_repo_url.startswith("http"):
+        full_repo_url = f"https://github.com/{repo_url_input.replace('sources/github/', '')}"
+        if not full_repo_url.endswith(".git"): full_repo_url += ".git"
+    
+    repo_url = transform_github_url(repo_url_input)
 
     hf_profile, hf_space, hf_token = hf_prof_fallback, hf_space_fallback, hf_token_fallback
     if log_file:
@@ -1159,7 +1185,9 @@ async def handle_jules_comm(repo_url, branch="main", log_file="", hf_prof_fallba
                 hf_token = get_hf_token(hf_profile) or hf_token_fallback
 
     base_instruction = (
-        "important!: The current codebase you are working in is to be adapted following the vision of the project transformation in the /jules folder. "
+        f"important!: You must first clone the repository into your current workspace using its full HTTPS link: {full_repo_url}. "
+        "Investigate the existing codebase to understand its structure and features. "
+        "The current codebase you are working in is to be adapted following the vision of the project transformation in the /jules folder. "
         "The current application should be kept and only slightly changed, improved, expanded by the features described in /jules folder. "
         "The tasks files give a starting point, but judge for yourself. Develop tests to run to first see the current application feature, how it is working and make a plan to implement the features requested in /jules folder to develop the current application inside the repo towards the vision described to achieve expansion of the current app with functionalities described in the /jules folder. "
         "Adapt your coding implemententation to the coding language used by the project, and try to go with that. Test the full app within this working space.\n\n"
@@ -1787,8 +1815,9 @@ with gr.Blocks() as demo:
         async def respond(message, chat_history):
             async for updated_history in handle_chat(message, chat_history, persona="planning"):
                 yield "", updated_history
-        submit_btn.click(respond, [msg, chatbot], [msg, chatbot])
-        msg.submit(respond, [msg, chatbot], [msg, chatbot])
+
+        p_click = submit_btn.click(respond, [msg, chatbot], [msg, chatbot])
+        p_submit = msg.submit(respond, [msg, chatbot], [msg, chatbot])
         plan_back_btn.click(lambda: (gr.update(visible=True), gr.update(visible=False)), None, [planning_chat_col, planning_structured_col])
 
     with gr.Tab("Push Files") as push_files_tab:
@@ -1880,8 +1909,7 @@ with gr.Blocks() as demo:
             res, sid = await handle_jules_comm(repo, branch, log_file, hf_prof, hf_space, hf_token)
             return res, sid
 
-        comm_btn.click(start_comm_and_monitor, [repo_input, branch_dropdown, log_dropdown_comm, hf_profile_comm, hf_space_comm, hf_token_comm], [comm_output, mon_session_id]) \
-                .then(handle_supervisor_nudge, inputs=[mon_session_id, log_dropdown_comm], outputs=[nudge_output])
+        comm_btn.click(start_comm_and_monitor, [repo_input, branch_dropdown, log_dropdown_comm, hf_profile_comm, hf_space_comm, hf_token_comm], [comm_output, mon_session_id])
 
     with gr.Tab("Container Logs"):
         log_view = gr.Textbox(label="app.log (last 100 lines)", lines=25, interactive=False)
@@ -1904,12 +1932,19 @@ with gr.Blocks() as demo:
         [mentor_chatbot], 
         [mentor_chat_col, mentor_structured_col, mentor_proj_desc, mentor_tasks_tests, mentor_repos, mentor_expectations, mentor_endpoints, mentor_prof, mentor_space, mentor_token, mentor_status, log_dropdown, log_dropdown_comm]
     )
+    
+    planning_outputs = [planning_chat_col, planning_structured_col, plan_proj_desc, plan_tasks_tests, plan_repos, plan_expectations, plan_endpoints, plan_prof, plan_space, plan_token, ideate_status, log_dropdown, log_dropdown_comm]
+    
     close_btn.click(
         planning_close_wrap, 
         [chatbot], 
-        [planning_chat_col, planning_structured_col, plan_proj_desc, plan_tasks_tests, plan_repos, plan_expectations, plan_endpoints, plan_prof, plan_space, plan_token, ideate_status, log_dropdown, log_dropdown_comm]
+        planning_outputs
     )
 
+    # Planning session auto-pipeline (trigger close after final answer)
+    p_click.then(planning_close_wrap, [chatbot], planning_outputs)
+    p_submit.then(planning_close_wrap, [chatbot], planning_outputs)
+
     # Log Refresh Logic
     refresh_logs_btn_1.click(refresh_logs_ui, outputs=[log_dropdown, log_dropdown_comm])
     refresh_logs_btn_2.click(refresh_logs_ui, outputs=[log_dropdown, log_dropdown_comm])

test_models.py

@@ -0,0 +1,49 @@
+import asyncio
+import os
+import json
+import re
+from unittest.mock import AsyncMock, MagicMock, patch
+import sys
+
+# Mock modules
+sys.modules['gradio'] = MagicMock()
+sys.modules['git'] = MagicMock()
+
+import app
+from langchain_core.messages import AIMessage
+
+async def test_persona_models():
+    # Reset model mocks if they were set
+    # In app.py:
+    # extraction_llm = large_llm if persona == "mentor" else fast_llm
+    
+    app.large_llm = MagicMock()
+    app.fast_llm = MagicMock()
+    app.get_ideation_logs = MagicMock(return_value=[])
+    
+    # Mock ainvoke_with_retry to check which LLM is called
+    called_llms = []
+    async def mock_invoke(llm, prompt, **kwargs):
+        called_llms.append(llm)
+        return AIMessage(content="Extracted")
+
+    with patch("app.ainvoke_with_retry", side_effect=mock_invoke), \
+         patch("os.makedirs"), patch("builtins.open", MagicMock()):
+        
+        # Mentor Persona
+        print("Testing Mentor extraction (should use large_llm)...")
+        await app.handle_close_ideate([{"role": "user", "content": "hi"}], persona="mentor")
+        # In handle_close_ideate, it makes 8 calls
+        assert all(l == app.large_llm for l in called_llms[:8])
+        
+        called_llms.clear()
+        
+        # Planning Persona
+        print("Testing Planning extraction (should use fast_llm)...")
+        await app.handle_close_ideate([{"role": "user", "content": "hi"}], persona="planning")
+        assert all(l == app.fast_llm for l in called_llms[:8])
+
+    print("Persona model selection test PASSED!")
+
+if __name__ == "__main__":
+    asyncio.run(test_persona_models())