fix api key management in backoffice

apps/vedana/.env.example

@@ -24,8 +24,11 @@ JIMS_DB_CONN_URI="postgresql://postgres:postgres@db:5432"
 
 # Main LLM model
 MODEL="gpt-4.1-mini"
-# Embeddings are provided via LiteLLM as well. OpenRouter's embedding models are not supported there at the moment
+# Embeddings are provided via LiteLLM as well.
+# You can use OpenAI directly:
 EMBEDDINGS_MODEL="text-embedding-3-large"
+# Or route embeddings through OpenRouter (requires OPENROUTER_API_KEY):
+# EMBEDDINGS_MODEL="openrouter/openai/text-embedding-3-large"
 # Note: changing dimensions of embeddings requires an SQL migration.
 EMBEDDINGS_DIM=1024
 EMBEDDINGS_MAX_BATCH_SIZE=2048

libs/jims-core/src/jims_core/llms/llm_provider.py

@@ -24,13 +24,6 @@ class LLMSettings(BaseSettings):
     embeddings_max_batch_size: int = 2048
     embeddings_max_tokens_per_batch: int = 200000
 
-    # passable api_keys; if None, defaults to env vars
-    model_api_key: str | None = None
-    embeddings_model_api_key: str | None = None
-
-    # openrouter_api_key: str | None = None
-    openrouter_api_base_url: str = "https://openrouter.ai/api/v1"
-
 
 env_settings = LLMSettings()  # type: ignore
 
@@ -77,9 +70,7 @@ class LLMProvider:
     def __init__(self, settings: LLMSettings | None = None) -> None:
         self._settings = settings or env_settings
         self.model = self._settings.model
-        self.model_api_key = self._settings.model_api_key
         self.embeddings_model = self._settings.embeddings_model
-        self.embeddings_model_api_key = self._settings.embeddings_model_api_key
         self.embeddings_dim = self._settings.embeddings_dim
         self.max_batch_size = self._settings.embeddings_max_batch_size
         self.max_tokens_per_batch = self._settings.embeddings_max_tokens_per_batch
@@ -109,7 +100,7 @@ def observe_completion(self, completion: litellm.ModelResponse) -> None:
             prompt_tokens=prompt_tokens,
             completion_tokens=completion_tokens,
             cached_tokens=cached_tokens,
-            request_cost=completion._hidden_params.get("response_cost", 0),  # or litellm.completion_cost(completion)
+            request_cost=completion._hidden_params.get("response_cost") or 0,  # or litellm.completion_cost(completion)
         )
 
         llm_calls_total.labels(completion.model).inc()
@@ -125,7 +116,7 @@ def observe_create_embedding(self, res: litellm.EmbeddingResponse) -> None:
         if usage is not None:
             self.usage[model].observe(
                 prompt_tokens=usage.prompt_tokens,
-                request_cost=res._hidden_params.get("response_cost", 0),  # or litellm.completion_cost(res)
+                request_cost=res._hidden_params.get("response_cost") or 0,  # or litellm.completion_cost(res)
             )
             llm_usage_prompt_tokens_total.labels(model).inc(usage.prompt_tokens)
 
@@ -135,7 +126,6 @@ async def create_embedding(self, text: str) -> list[float]:
             model=self.embeddings_model,
             input=[text],
             dimensions=self.embeddings_dim,
-            api_key=self.embeddings_model_api_key,
         )
         self.observe_create_embedding(response)
         return response.data[0]["embedding"]
@@ -170,7 +160,6 @@ async def create_embeddings(self, texts: list[str]) -> list[list[float]]:
                 model=self.embeddings_model,
                 input=batch,
                 dimensions=self.embeddings_dim,
-                api_key=self.embeddings_model_api_key,
             )
             self.observe_create_embedding(response)
             results.extend(d["embedding"] for d in response.data)
@@ -181,7 +170,6 @@ def create_embedding_sync(self, text: str) -> list[float]:
             model=self.embeddings_model,
             input=[text],
             dimensions=self.embeddings_dim,
-            api_key=self.embeddings_model_api_key,
         )
         self.observe_create_embedding(response)
         return response.data[0]["embedding"]
@@ -194,7 +182,6 @@ def create_embeddings_sync(self, texts: list[str]) -> list[list[float]]:
                 model=self.embeddings_model,
                 input=batch,
                 dimensions=self.embeddings_dim,
-                api_key=self.embeddings_model_api_key,
             )
             self.observe_create_embedding(response)
             results.extend(d["embedding"] for d in response.data)
@@ -210,7 +197,6 @@ async def chat_completion_structured[T: BaseModel](
             model=self.model,
             messages=list(messages),
             response_format=response_format,
-            api_key=self.model_api_key,
         )
         assert isinstance(completion, litellm.ModelResponse)
 
@@ -234,7 +220,6 @@ async def chat_completion_plain(
             model=self.model,
             messages=list(messages),
             caching=use_cache,
-            api_key=self.model_api_key,
         )
         assert isinstance(completion, litellm.ModelResponse)
 
@@ -258,7 +243,6 @@ async def chat_completion_with_tools(
             model=self.model,
             messages=list(messages),
             tools=tools,
-            api_key=self.model_api_key,
         )
         assert isinstance(completion, litellm.ModelResponse)
 

libs/jims-core/src/jims_core/thread/thread_context.py

@@ -111,7 +111,7 @@ def context(self, conversation_length: int = 20) -> list[CommunicationEvent]:
                 result.append(CommunicationEvent(**event.event_data))
                 comm_counter += 1
             if event.event_type.startswith("context."):
-                result.append(CommunicationEvent(**event.event_data))    
+                result.append(CommunicationEvent(**event.event_data))
             if comm_counter > conversation_length:
                 break
 

libs/vedana-backoffice/pyproject.toml

@@ -10,7 +10,7 @@ authors = [
 ]
 
 dependencies = [
-    "reflex>=0.8.26,<0.9.0",
+    "reflex>=0.8.27,<0.9.0",
     "orjson>=3.11.3",
     "vedana-core",
     "vedana-etl",

libs/vedana-backoffice/src/vedana_backoffice/pages/chat.py

@@ -2,7 +2,7 @@
 
 from vedana_backoffice.components.ui_chat import render_message_bubble
 from vedana_backoffice.states.chat import ChatState
-from vedana_backoffice.states.common import AppVersionState
+from vedana_backoffice.states.common import AppVersionState, DebugState
 from vedana_backoffice.ui import app_header
 
 
@@ -111,7 +111,7 @@ def page() -> rx.Component:
                     rx.cond(
                         ChatState.enable_dm_filtering & AppVersionState.debug_mode,
                         rx.select(
-                            items=ChatState.dm_filter_available_models,
+                            items=DebugState.available_models,
                             value=ChatState.dm_filter_model,
                             on_change=ChatState.set_dm_filter_model,
                             width="16em",
@@ -120,6 +120,23 @@ def page() -> rx.Component:
                         rx.text(ChatState.dm_filter_model, size="1", color="gray"),
                     ),
                     rx.spacer(),
+                    rx.cond(
+                        AppVersionState.debug_mode,
+                        rx.cond(
+                            DebugState.embeddings_model_available,
+                            rx.text(
+                                f"Embeddings: {DebugState.embeddings_model}",
+                                size="1",
+                                color="gray",
+                            ),
+                            rx.text(
+                                f"Embeddings: {DebugState.default_embeddings_model} (unavailable for provider)",
+                                size="1",
+                                color="red",
+                            ),
+                        ),
+                        rx.fragment(),
+                    ),
                     rx.hstack(
                         rx.cond(
                             ChatState.chat_thread_id != "",
@@ -157,21 +174,12 @@ def page() -> rx.Component:
                         ),
                         rx.cond(
                             ChatState.model_selection_allowed,
-                            rx.hstack(
-                                rx.select(
-                                items=["openai", "openrouter"],
-                                value=ChatState.provider,
-                                on_change=ChatState.set_provider,
-                                width="10em",
-                                placeholder="Provider",
-                                ),
-                                rx.select(
-                                    items=ChatState.available_models,
-                                    value=ChatState.model,
-                                    on_change=ChatState.set_model,
-                                    width="16em",
-                                    placeholder="Select model",
-                                ),
+                            rx.select(
+                                items=DebugState.available_models,
+                                value=ChatState.model,
+                                on_change=ChatState.set_model,
+                                width="20em",
+                                placeholder="Select model",
                             ),
                             rx.badge(ChatState.model, variant="surface", color_scheme="gray", size="3"),
                         ),

libs/vedana-backoffice/src/vedana_backoffice/pages/eval.py

@@ -1,8 +1,8 @@
 import reflex as rx
 
-from vedana_backoffice.states.common import AppVersionState
-from vedana_backoffice.states.eval import EvalState, RunSummary
 from vedana_backoffice.states.chat import ChatState
+from vedana_backoffice.states.common import AppVersionState, DebugState
+from vedana_backoffice.states.eval import EvalState, RunSummary
 from vedana_backoffice.ui import app_header
 
 
@@ -21,7 +21,11 @@ def _selection_and_actions() -> rx.Component:
                 color_scheme="blue",
                 on_click=EvalState.run_selected_tests,
                 loading=EvalState.is_running,
-                disabled=rx.cond(EvalState.can_run, False, True),  # type: ignore[arg-type]
+                disabled=rx.cond(
+                    DebugState.debug_mode & ~DebugState.embeddings_model_available,
+                    True,
+                    rx.cond((EvalState.selected_count > 0) & ~EvalState.is_running, False, True),  # type: ignore[arg-type]
+                ),
             ),
             rx.spacer(),
             rx.button(
@@ -175,7 +179,7 @@ def _judge_card() -> rx.Component:
                 rx.cond(
                     AppVersionState.debug_mode,
                     rx.select(
-                        items=EvalState.judge_available_models,
+                        items=DebugState.available_models,
                         value=EvalState.judge_model,
                         on_change=EvalState.set_judge_model,
                         width="100%",
@@ -235,25 +239,12 @@ def _pipeline_card() -> rx.Component:
                     rx.text("Pipeline model", weight="medium", width="100%"),
                     rx.cond(
                         ChatState.model_selection_allowed,
-                        rx.hstack(
-                            rx.select(
-                                items=["openai", "openrouter"],
-                                value=EvalState.provider,
-                                on_change=EvalState.set_provider,
-                                width="100%",
-                                placeholder="Provider",
-                            ),
-                            rx.select(
-                                items=EvalState.available_models_view,
-                                value=EvalState.pipeline_model,
-                                on_change=EvalState.set_pipeline_model,
-                                width="100%",
-                                placeholder="Select model",
-                            ),
-                            spacing="2",
-                            align="center",
-                            wrap="wrap",
+                        rx.select(
+                            items=DebugState.available_models,
+                            value=EvalState.pipeline_model,
+                            on_change=EvalState.set_pipeline_model,
                             width="100%",
+                            placeholder="Select model",
                         ),
                         rx.text(EvalState.pipeline_model, size="3"),
                     ),
@@ -270,7 +261,7 @@ def _pipeline_card() -> rx.Component:
                             rx.cond(
                                 AppVersionState.debug_mode,
                                 rx.select(
-                                    items=EvalState.dm_filter_available_models,
+                                    items=DebugState.available_models,
                                     value=EvalState.dm_filter_model,
                                     on_change=EvalState.set_dm_filter_model,
                                     width="100%",
@@ -291,9 +282,18 @@ def _pipeline_card() -> rx.Component:
                 ),
                 rx.box(
                     rx.text("Embeddings", weight="medium"),
-                    rx.text(
-                        rx.cond(EvalState.embeddings_model != "", EvalState.embeddings_model, "—"),
-                        size="3",
+                    rx.cond(
+                        AppVersionState.debug_mode,
+                        rx.cond(
+                            DebugState.embeddings_model_available,
+                            rx.text(DebugState.embeddings_model, size="3"),
+                            rx.text(
+                                f"{EvalState.default_embeddings_model} (unavailable for provider)",
+                                size="3",
+                                color="red",
+                            ),
+                        ),
+                        rx.text(EvalState.default_embeddings_model, size="3"),
                     ),
                     rx.text(
                         EvalState.embeddings_dim_label,

libs/vedana-backoffice/src/vedana_backoffice/pages/jims_thread_list_page.py

@@ -687,7 +687,13 @@ def priority_badge(value: str) -> rx.Component:  # type: ignore[valid-type]
         rx.text(ThreadListState.rows_display, size="2", color="gray"),
         rx.spacer(),
         rx.hstack(
-            rx.button("⏮", variant="soft", size="1", on_click=ThreadListState.first_page, disabled=~ThreadListState.has_prev_page),  # type: ignore[operator]
+            rx.button(
+                "⏮",
+                variant="soft",
+                size="1",
+                on_click=ThreadListState.first_page,
+                disabled=~ThreadListState.has_prev_page,
+            ),  # type: ignore[operator]
             rx.button(
                 "← Prev",
                 variant="soft",
@@ -703,7 +709,13 @@ def priority_badge(value: str) -> rx.Component:  # type: ignore[valid-type]
                 on_click=ThreadListState.next_page,
                 disabled=~ThreadListState.has_next_page,
             ),  # type: ignore[operator]
-            rx.button("⏭", variant="soft", size="1", on_click=ThreadListState.last_page, disabled=~ThreadListState.has_next_page),  # type: ignore[operator]
+            rx.button(
+                "⏭",
+                variant="soft",
+                size="1",
+                on_click=ThreadListState.last_page,
+                disabled=~ThreadListState.has_next_page,
+            ),  # type: ignore[operator]
             spacing="2",
             align="center",
         ),