Upload folder using huggingface_hub

deep_research.py

@@ -7,7 +7,7 @@ import re
 load_dotenv(override=True)
 
 
-async def run_research(query: str, progress=gr.Progress()):
+async def run_research(query: str, model_choice: str, progress=gr.Progress()):
     """Run research and yield updates for both report and references"""
     status_messages = []
     final_report_md = ""
@@ -39,7 +39,7 @@ async def run_research(query: str, progress=gr.Progress()):
     '''
 
     # Collect all chunks and parse structured messages
-    async for chunk in ResearchManager().run(query):
+    async for chunk in ResearchManager(model_choice).run(query):
         # Parse structured messages (format: TYPE|data)
         if "|" in chunk:
             msg_type, msg_data = chunk.split("|", 1)
@@ -468,11 +468,36 @@ textarea:focus {
     font-size: 1.125rem;
 }
 
-/* Primary Run Button - Minimal & Modern */
+/* Input Row with Button and Dropdown */
 .button-row-bottom {
     display: flex;
     gap: 0.75rem;
     margin-top: 1rem;
+    align-items: center;
+}
+
+/* Model Selector Dropdown */
+.model-selector select {
+    background: rgba(45, 45, 45, 0.6) !important;
+    border: 1px solid rgba(55, 65, 81, 0.5) !important;
+    color: #9CA3AF !important;
+    border-radius: 10px !important;
+    padding: 0.875rem 1rem !important;
+    font-size: 0.875rem !important;
+    font-weight: 500 !important;
+    transition: all 0.2s ease !important;
+    cursor: pointer !important;
+}
+
+.model-selector select:hover {
+    border-color: rgba(16, 185, 129, 0.4) !important;
+    background: rgba(45, 45, 45, 0.8) !important;
+}
+
+.model-selector select:focus {
+    border-color: rgba(16, 185, 129, 0.6) !important;
+    box-shadow: 0 0 0 3px rgba(16, 185, 129, 0.1) !important;
+    outline: none !important;
 }
 
 button[variant="primary"] {
@@ -762,7 +787,24 @@ with gr.Blocks(theme=luntre_theme, css=custom_css, title="Luntre AI - Deep Resea
         )
 
         with gr.Row(elem_classes="button-row-bottom"):
-            run_btn = gr.Button("🚀 Run Research", variant="primary", scale=1)
+            run_btn = gr.Button("🚀 Run Research", variant="primary", scale=2)
+
+            model_selector = gr.Dropdown(
+                choices=[
+                    "gemini-2.5-flash (Default)",
+                    "gemini-2.0-flash-exp",
+                    "gemini-2.0-flash-thinking-exp",
+                    "llama-3.3-70b-versatile (Groq)",
+                ],
+                value="gemini-2.5-flash (Default)",
+                label="",
+                show_label=False,
+                container=False,
+                elem_classes="model-selector",
+                scale=1,
+                interactive=True
+            )
+
             # Hidden buttons for programmatic access
             edit_btn = gr.Button("Edit", variant="secondary", visible=False, elem_id="edit-query-btn")
             rewrite_btn = gr.Button("Rewrite", variant="secondary", visible=False, elem_id="rewrite-btn")
@@ -793,7 +835,7 @@ with gr.Blocks(theme=luntre_theme, css=custom_css, title="Luntre AI - Deep Resea
         queue=False
     ).then(
         fn=run_research,
-        inputs=[current_query_state],
+        inputs=[current_query_state, model_selector],
         outputs=[report_output, references_output]
     )
 
@@ -808,7 +850,7 @@ with gr.Blocks(theme=luntre_theme, css=custom_css, title="Luntre AI - Deep Resea
     # Rewrite (run again with same query)
     rewrite_event = rewrite_btn.click(
         fn=run_research,
-        inputs=[current_query_state],
+        inputs=[current_query_state, model_selector],
         outputs=[report_output, references_output]
     )
 
@@ -825,7 +867,7 @@ with gr.Blocks(theme=luntre_theme, css=custom_css, title="Luntre AI - Deep Resea
         queue=False
     ).then(
         fn=run_research,
-        inputs=[current_query_state],
+        inputs=[current_query_state, model_selector],
         outputs=[report_output, references_output]
     )
 

model_config.py

@@ -0,0 +1,66 @@
+import os
+from dotenv import load_dotenv
+from agents import AsyncOpenAI, OpenAIChatCompletionsModel
+
+load_dotenv(override=True)
+
+
+def get_model(model_choice: str):
+    """
+    Get the appropriate model based on user selection
+
+    Args:
+        model_choice: String from dropdown (e.g., "gemini-2.5-flash (Default)")
+
+    Returns:
+        OpenAIChatCompletionsModel configured for the selected model
+    """
+    # Extract model name from dropdown choice
+    if "gemini-2.5-flash" in model_choice.lower():
+        model_name = "gemini-2.5-flash"
+        api_key = os.getenv('GEMINI_API_KEY')
+        base_url = "https://generativelanguage.googleapis.com/v1beta/openai/"
+
+    elif "gemini-2.0-flash-thinking" in model_choice.lower():
+        model_name = "gemini-2.0-flash-thinking-exp-01-21"
+        api_key = os.getenv('GEMINI_API_KEY')
+        base_url = "https://generativelanguage.googleapis.com/v1beta/openai/"
+
+    elif "gemini-2.0-flash-exp" in model_choice.lower():
+        model_name = "gemini-2.0-flash-exp"
+        api_key = os.getenv('GEMINI_API_KEY')
+        base_url = "https://generativelanguage.googleapis.com/v1beta/openai/"
+
+    elif "llama" in model_choice.lower():
+        model_name = "llama-3.3-70b-versatile"
+        api_key = os.getenv('GROQ_API_KEY')
+        base_url = "https://api.groq.com/openai/v1"
+
+    else:
+        # Default to gemini-2.5-flash
+        model_name = "gemini-2.5-flash"
+        api_key = os.getenv('GEMINI_API_KEY')
+        base_url = "https://generativelanguage.googleapis.com/v1beta/openai/"
+
+    # Create client
+    client = AsyncOpenAI(
+        api_key=api_key,
+        base_url=base_url
+    )
+
+    # Return model
+    return OpenAIChatCompletionsModel(
+        model=model_name,
+        openai_client=client
+    )
+
+
+def get_model_display_name(model_choice: str) -> str:
+    """Get a clean display name for the model"""
+    model_map = {
+        "gemini-2.5-flash (Default)": "Gemini 2.5 Flash",
+        "gemini-2.0-flash-exp": "Gemini 2.0 Flash",
+        "gemini-2.0-flash-thinking-exp": "Gemini 2.0 Flash Thinking",
+        "llama-3.3-70b-versatile (Groq)": "Llama 3.3 70B (Groq)"
+    }
+    return model_map.get(model_choice, model_choice)

research_manager.py

@@ -1,18 +1,27 @@
-from agents import Runner, trace, gen_trace_id
+from agents import Runner, trace, gen_trace_id, Agent, ModelSettings
 from search_agent import search_agent
 from planner_agent import planner_agent, WebSearchItem, WebSearchPlan
-from writer_agent import writer_agent_plain, writer_agent, ReportData
+from writer_agent import writer_agent_plain, writer_agent, ReportData, INSTRUCTIONS as WRITER_INSTRUCTIONS
 from email_agent import email_agent
 import asyncio
 import time
 from pydantic import BaseModel, Field
+from model_config import get_model, get_model_display_name
 
 class ResearchManager:
-    def __init__(self):
+    def __init__(self, model_choice="gemini-2.5-flash (Default)"):
         # Track request timestamps for intelligent rate limiting
         self.request_times = []
-        self.max_rpm = 10  # Gemini 2.5 Flash free tier: 10 requests per minute
+        self.max_rpm = 10  # Default: 10 requests per minute
         self.rate_limit_window = 60  # seconds
+        self.model_choice = model_choice
+        self.model = get_model(model_choice)
+
+        # Adjust rate limits based on model
+        if "groq" in model_choice.lower():
+            self.max_rpm = 30  # Groq has higher limits
+        elif "thinking" in model_choice.lower():
+            self.max_rpm = 5  # Experimental models have lower limits
 
     async def wait_for_rate_limit(self):
         """Intelligent rate limiting: only sleep as much as needed"""
@@ -36,15 +45,11 @@ class ResearchManager:
     async def run(self, query: str):
         """ Run the deep research process, yielding the status updates and the final report"""
         trace_id = gen_trace_id()
-
-        # Closure to yield progress updates
-        async def yield_progress(message):
-            # This is a workaround - we'll store messages in a list
-            pass
+        model_display = get_model_display_name(self.model_choice)
 
         with trace("Research trace", trace_id=trace_id):
-            print(f"Using Brave Search API and Gemini 2.5 Flash")
-            yield f"INIT|Using Brave Search API and Gemini 2.5 Flash"
+            print(f"Using Brave Search API and {model_display}")
+            yield f"INIT|Using Brave Search API and {model_display}"
             print("Starting research...")
 
             search_plan = await self.plan_searches(query)
@@ -78,8 +83,18 @@ class ResearchManager:
         """ Plan the searches to perform for the query """
         print("Planning searches...")
         await self.wait_for_rate_limit()
+
+        # Use selected model for planning
+        from planner_agent import INSTRUCTIONS as PLANNER_INSTRUCTIONS, WebSearchPlan
+        dynamic_planner = Agent(
+            name="PlannerAgent",
+            instructions=PLANNER_INSTRUCTIONS,
+            model=self.model,
+            output_type=WebSearchPlan,
+        )
+
         result = await Runner.run(
-            planner_agent,
+            dynamic_planner,
             f"Query: {query}",
         )
         print(f"Will perform {len(result.final_output.searches)} searches")
@@ -140,6 +155,18 @@ class ResearchManager:
         print("Thinking about report...")
         input = f"Original query: {query}\nSummarized search results: {search_results}"
 
+        # Create writer with selected model
+        dynamic_writer = Agent(
+            name="WriterAgentPlain",
+            instructions=WRITER_INSTRUCTIONS,
+            model=self.model,
+            output_type=str,
+            model_settings=ModelSettings(
+                max_tokens=32000,
+                temperature=0.7,
+            ),
+        )
+
         max_retries = 3
         for attempt in range(max_retries):
             try:
@@ -147,7 +174,7 @@ class ResearchManager:
                 await self.wait_for_rate_limit()
                 # Use plain text agent to avoid structured output truncation
                 result = await Runner.run(
-                    writer_agent_plain,
+                    dynamic_writer,
                     input,
                 )
                 print("Finished writing report")