Improve audio capture and LLM integration

- Implement Silero VAD for dynamic audio chunking
- Add support for Ollama and vLLM backends
- Harden extraction prompts for strict JSON output
- Refactor TUI worker to handle proposals asynchronously

src/llm/processor.py

@@ -22,10 +22,34 @@ class LLMProcessor:
         :param base_url: OpenAI-compatible base URL (e.g., for vLLM).
         :param model: The model to use for processing. If None, it looks for LLM_MODEL in environment variables.
         """
-        self.client = OpenAI(
-            api_key=api_key or os.environ.get("OPENAI_API_KEY"),
-            base_url=base_url or os.environ.get("OPENAI_BASE_URL"),
-        )
+        backend = os.environ.get("LLM_BACKEND", "openai").lower()
+
+        if backend == "ollama":
+            # Ollama's OpenAI-compatible API
+            final_base_url = base_url or "http://localhost:11434/v1"
+            final_api_key = api_key or "ollama"
+        elif backend == "vllm":
+            # Remote vLLM server
+            final_base_url = base_url or os.environ.get("OPENAI_BASE_URL")
+            final_api_key = api_key or os.environ.get("OPENAI_API_KEY")
+        else:  # default to openai
+            final_base_url = base_url or os.environ.get("OPENAI_BASE_URL")
+            final_api_key = api_key or os.environ.get("OPENAI_API_KEY")
+
+        try:
+            self.client = OpenAI(
+                api_key=final_api_key,
+                base_url=final_base_url,
+            )
+            # Simple connectivity check for local backends
+            if backend == "ollama":
+                # We can't easily check connectivity without making a call,
+                # but we can ensure the client is initialized.
+                pass
+        except Exception as e:
+            print(f"Error initializing LLM client for backend {backend}: {e}")
+            raise
+
         self.model = model or os.environ.get("LLM_MODEL", "gpt-4o")
 
     def _call_llm(
@@ -67,6 +91,7 @@ class LLMProcessor:
         print(f"LLM Processor (Extract): Calling extraction for: {filtered_text}")
         try:
             # Using standard chat.completions.create with JSON mode for better compatibility with vLLM
+            print("LLM Processor (Extract): Sending request to backend...")
             response = self.client.chat.completions.create(
                 model=self.model,
                 messages=[
@@ -76,6 +101,7 @@ class LLMProcessor:
                 response_format={"type": "json_object"},
                 extra_body={"include_reasoning": False},
             )
+            print("LLM Processor (Extract): Response received from backend.")
 
             import json