add model selection backend support

backend/requirements.txt

@@ -14,4 +14,5 @@ openai==2.36.0
 google-auth==2.0.0
 requests==2.32.3
 python-multipart==0.0.9
-PyYAML==6.0.1
+PyYAML==6.0.1
+cachetools==5.5.0

backend/src/dna/llm_providers/gemini_provider.py

@@ -3,11 +3,16 @@
 Gemini implementation of the LLM provider interface.
 """
 
+import logging
 import os
+from typing import Any
 
+from cachetools import TTLCache
 from openai import AsyncOpenAI
 
-from dna.llm_providers.llm_provider_base import LLMProviderBase
+from dna.llm_providers.llm_provider_base import MODEL_CACHE_TTL, LLMProviderBase
+
+logger = logging.getLogger(__name__)
 
 
 class GeminiProvider(LLMProviderBase):
@@ -18,10 +23,35 @@ class GeminiProvider(LLMProviderBase):
     DEFAULT_MODEL = "gemini-2.5-flash"
     DEFAULT_URL = "https://generativelanguage.googleapis.com/v1beta/openai/"
 
+    def __init__(self, **kwargs):
+        super().__init__(**kwargs)
+        self._models_cache: TTLCache = TTLCache(maxsize=1, ttl=MODEL_CACHE_TTL)
+
     def _get_provider_client(self):
         """Construct an instance of the LLM provider's client."""
         return AsyncOpenAI(
             api_key=self.api_key,
             base_url=os.getenv(f"{self.LLM_PROVIDER_NAME }_URL", self.DEFAULT_URL),
             timeout=self.timeout,
         )
+
+    async def get_available_models(self) -> dict[str, Any]:
+        """Fetch available models from Gemini API with caching."""
+        cache_key = "models"
+        if cache_key in self._models_cache:
+            return self._models_cache[cache_key]
+
+        try:
+            response = await self.client.models.list()
+            model_ids = sorted(m.id for m in response.data)
+        except Exception:
+            logger.warning("Failed to fetch models from Gemini API, using default")
+            model_ids = [self.model]
+
+        result = {
+            "provider": "gemini",
+            "models": model_ids,
+            "default": self.model,
+        }
+        self._models_cache[cache_key] = result
+        return result

backend/src/dna/llm_providers/llm_provider_base.py

@@ -14,6 +14,8 @@
 
 from dna.prompts.generate_note_prompt import GENERATE_NOTE_PROMPT
 
+MODEL_CACHE_TTL = 3600  # 1 hour
+
 logger = logging.getLogger(__name__)
 
 T = TypeVar("T", bound=BaseModel)
@@ -161,13 +163,26 @@ async def close(self) -> None:
             await self._client.close()
             self._client = None
 
+    async def get_available_models(self) -> dict[str, Any]:
+        """Return available models for this provider.
+
+        Returns a dict with keys: provider, models, default.
+        Subclasses should override to provide dynamic discovery with caching.
+        """
+        return {
+            "provider": (self.LLM_PROVIDER_NAME or "").lower(),
+            "models": [self.model],
+            "default": self.model,
+        }
+
     async def generate_note(
         self,
         prompt: str,
         transcript: str,
         context: str,
         existing_notes: str,
         additional_instructions: Optional[str] = None,
+        model: Optional[str] = None,
     ) -> str:
         """Generate a note suggestion from the given inputs.
 
@@ -177,10 +192,13 @@ async def generate_note(
             context: Version context (entity name, task, status, etc.).
             existing_notes: Any notes the user has already written.
             additional_instructions: Optional additional instructions to append.
+            model: Optional model override; falls back to self.model.
 
         Returns:
             The generated note suggestion.
         """
+        use_model = model or self.model
+
         user_message = self._substitute_template(
             prompt, transcript, context, existing_notes
         )
@@ -189,7 +207,7 @@ async def generate_note(
             user_message += f"\n\nAdditional Instructions: {additional_instructions}"
 
         response = await self.client.chat.completions.create(
-            model=self.model,
+            model=use_model,
             messages=[
                 {"role": "system", "content": GENERATE_NOTE_PROMPT},
                 {"role": "user", "content": user_message},

backend/src/dna/llm_providers/openai_provider.py

@@ -3,9 +3,17 @@
 OpenAI implementation of the LLM provider interface.
 """
 
+import logging
+from typing import Any
+
+from cachetools import TTLCache
 from openai import AsyncOpenAI
 
-from dna.llm_providers.llm_provider_base import LLMProviderBase
+from dna.llm_providers.llm_provider_base import MODEL_CACHE_TTL, LLMProviderBase
+
+logger = logging.getLogger(__name__)
+
+OPENAI_CHAT_PREFIXES = ("gpt-", "o1", "o3", "o4", "chatgpt-")
 
 
 class OpenAIProvider(LLMProviderBase):
@@ -15,6 +23,35 @@ class OpenAIProvider(LLMProviderBase):
 
     DEFAULT_MODEL = "gpt-4o-mini"
 
+    def __init__(self, **kwargs):
+        super().__init__(**kwargs)
+        self._models_cache: TTLCache = TTLCache(maxsize=1, ttl=MODEL_CACHE_TTL)
+
     def _get_provider_client(self):
         """Construct an instance of the LLM provider's client."""
         return AsyncOpenAI(api_key=self.api_key, timeout=self.timeout)
+
+    async def get_available_models(self) -> dict[str, Any]:
+        """Fetch available chat-completion models from OpenAI API with caching."""
+        cache_key = "models"
+        if cache_key in self._models_cache:
+            return self._models_cache[cache_key]
+
+        try:
+            response = await self.client.models.list()
+            model_ids = sorted(
+                m.id
+                for m in response.data
+                if m.id.startswith(OPENAI_CHAT_PREFIXES)
+            )
+        except Exception:
+            logger.warning("Failed to fetch models from OpenAI API, using default")
+            model_ids = [self.model]
+
+        result = {
+            "provider": "openai",
+            "models": model_ids,
+            "default": self.model,
+        }
+        self._models_cache[cache_key] = result
+        return result

backend/src/dna/models/requests.py

@@ -54,6 +54,10 @@ class GenerateNoteRequest(BaseModel):
         default=None,
         description="Optional additional instructions to append to the prompt",
     )
+    model: Optional[str] = Field(
+        default=None,
+        description="Optional LLM model override; omit to use server default",
+    )
 
 
 class GenerateNoteResponse(BaseModel):

backend/src/dna/models/user_settings.py

@@ -15,6 +15,10 @@ class UserSettingsUpdate(BaseModel):
     note_prompt: Optional[str] = Field(
         default=None, description="Custom prompt for generating notes"
     )
+    preferred_model: Optional[str] = Field(
+        default=None,
+        description="Preferred LLM model for note generation; empty means use server default",
+    )
     regenerate_on_version_change: Optional[bool] = Field(
         default=None,
         description="Regenerate AI note when switching review versions",
@@ -38,6 +42,7 @@ class UserSettings(BaseModel):
     id: str = Field(alias="_id")
     user_email: str
     note_prompt: str = ""
+    preferred_model: str = ""
     regenerate_on_version_change: bool = False
     regenerate_on_transcript_update: bool = False
     sync_prodtrack_tab_on_version_change: bool = True

backend/src/dna/models/user_settings_response.py

@@ -13,6 +13,7 @@ class UserSettingsResponse(BaseModel):
     id: str = Field(alias="_id")
     user_email: str
     note_prompt: str = ""
+    preferred_model: str = ""
     default_note_prompt: str = ""
     regenerate_on_version_change: bool = False
     regenerate_on_transcript_update: bool = False

backend/src/main.py

@@ -1418,6 +1418,7 @@ def _user_settings_to_response(settings: UserSettings) -> UserSettingsResponse:
         _id=settings.id,
         user_email=settings.user_email,
         note_prompt=settings.note_prompt,
+        preferred_model=settings.preferred_model,
         default_note_prompt=get_default_note_prompt(),
         regenerate_on_version_change=settings.regenerate_on_version_change,
         regenerate_on_transcript_update=settings.regenerate_on_transcript_update,
@@ -1438,6 +1439,7 @@ def _empty_user_settings_response(user_email: str) -> UserSettingsResponse:
         _id="",
         user_email=user_email,
         note_prompt="",
+        preferred_model="",
         default_note_prompt=default,
         regenerate_on_version_change=False,
         regenerate_on_transcript_update=False,
@@ -1787,6 +1789,20 @@ async def get_segments_for_version(
 # -----------------------------------------------------------------------------
 
 
+@app.get(
+    "/models",
+    tags=["LLM"],
+    summary="Get available LLM models",
+    description="Returns the list of models available from the active LLM provider.",
+)
+async def get_available_models(
+    llm_provider: LLMProviderDep,
+    _: CurrentUserDep,
+) -> dict:
+    """Get available models from the active LLM provider."""
+    return await llm_provider.get_available_models()
+
+
 def _build_full_prompt(
     prompt: str,
     transcript: str,
@@ -1852,12 +1868,17 @@ async def generate_note(
             prompt, transcript, context, existing_notes, request.additional_instructions
         )
 
+        model_override = request.model
+        if not model_override and user_settings:
+            model_override = user_settings.preferred_model or None
+
         suggestion = await llm_provider.generate_note(
             prompt=prompt,
             transcript=transcript,
             context=context,
             existing_notes=existing_notes,
             additional_instructions=request.additional_instructions,
+            model=model_override,
         )
 
         return GenerateNoteResponse(