fix: remove unnecessary tracker guards, forward variables and default_ai_provider

- Remove if-tracker guards in both runners since create_tracker is
  always set on enabled graphs (disabled graphs are filtered before
  runner creation), also fixing token_usage NameError when tracker=None
- Forward variables through _build_evaluator to _initialize_judges so
  judge templates can interpolate user-provided variables
- Add default_ai_provider param to agent_graph() and forward it to
  __evaluate_agent so graph node evaluators use the correct provider;
  propagate from create_agent_graph() as well

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Author: jsonbailey

packages/ai-providers/server-ai-langchain/src/ldai_langchain/langgraph_agent_graph_runner.py

@@ -304,7 +304,7 @@ async def run(self, input: Any) -> AgentGraphResult:
         """
         pending_eval_tasks: Dict[str, List[asyncio.Task]] = {}
         token = _run_eval_tasks.set(pending_eval_tasks)
-        tracker = self._graph.create_tracker() if self._graph.create_tracker is not None else None
+        tracker = self._graph.create_tracker()
         start_ns = time.perf_counter_ns()
 
         try:
@@ -325,12 +325,10 @@ async def run(self, input: Any) -> AgentGraphResult:
             # Flush per-node metrics to LD trackers
             await handler.flush(self._graph, pending_eval_tasks)
 
-            # Graph-level metrics
-            if tracker:
-                tracker.track_path(handler.path)
-                tracker.track_duration(duration)
-                tracker.track_invocation_success()
-                tracker.track_total_tokens(sum_token_usage_from_messages(messages))
+            tracker.track_path(handler.path)
+            tracker.track_duration(duration)
+            tracker.track_invocation_success()
+            tracker.track_total_tokens(sum_token_usage_from_messages(messages))
 
             return AgentGraphResult(
                 output=output,
@@ -347,9 +345,8 @@ async def run(self, input: Any) -> AgentGraphResult:
             else:
                 log.warning(f'LangGraphAgentGraphRunner run failed: {exc}')
             duration = (time.perf_counter_ns() - start_ns) // 1_000_000
-            if tracker:
-                tracker.track_duration(duration)
-                tracker.track_invocation_failure()
+            tracker.track_duration(duration)
+            tracker.track_invocation_failure()
             return AgentGraphResult(
                 output='',
                 raw=None,

packages/ai-providers/server-ai-openai/src/ldai_openai/openai_agent_graph_runner.py

@@ -76,7 +76,7 @@ async def run(self, input: Any) -> AgentGraphResult:
         :param input: The string prompt to send to the agent graph
         :return: AgentGraphResult with the final output and metrics
         """
-        tracker = self._graph.create_tracker() if self._graph.create_tracker is not None else None
+        tracker = self._graph.create_tracker()
         path: List[str] = []
         root_node = self._graph.root()
         root_key = root_node.get_key() if root_node else ''
@@ -99,14 +99,13 @@ async def run(self, input: Any) -> AgentGraphResult:
             self._track_tool_calls(result)
 
             duration = (time.perf_counter_ns() - start_ns) // 1_000_000
+            token_usage = get_ai_usage_from_response(result)
 
-            if tracker:
-                tracker.track_path(path)
-                tracker.track_duration(duration)
-                tracker.track_invocation_success()
-                token_usage = get_ai_usage_from_response(result)
-                if token_usage is not None:
-                    tracker.track_total_tokens(token_usage)
+            tracker.track_path(path)
+            tracker.track_duration(duration)
+            tracker.track_invocation_success()
+            if token_usage is not None:
+                tracker.track_total_tokens(token_usage)
 
             return AgentGraphResult(
                 output=str(result.final_output),
@@ -122,9 +121,8 @@ async def run(self, input: Any) -> AgentGraphResult:
             else:
                 log.warning(f'OpenAIAgentGraphRunner run failed: {exc}')
             duration = (time.perf_counter_ns() - start_ns) // 1_000_000
-            if tracker:
-                tracker.track_duration(duration)
-                tracker.track_invocation_failure()
+            tracker.track_duration(duration)
+            tracker.track_invocation_failure()
             return AgentGraphResult(
                 output='',
                 raw=None,

packages/sdk/server-ai/src/ldai/client.py

@@ -95,7 +95,7 @@ def _completion_config(
             key, context, default.to_dict(), variables
         )
 
-        evaluator = self._build_evaluator(judge_configuration, context, default_ai_provider)
+        evaluator = self._build_evaluator(judge_configuration, context, default_ai_provider, variables)
 
         config = AICompletionConfig(
             key=key,
@@ -327,20 +327,23 @@ def _build_evaluator(
         judge_configuration: Optional[JudgeConfiguration],
         context: Context,
         default_ai_provider: Optional[str] = None,
+        variables: Optional[Dict[str, Any]] = None,
     ) -> Evaluator:
         """
         Build an Evaluator for the given judge configuration.
 
         :param judge_configuration: The judge configuration listing judges to initialize
         :param context: Standard Context used when evaluating flags
         :param default_ai_provider: Optional default AI provider to use
+        :param variables: Optional variables for judge instruction interpolation
         :return: Evaluator wrapping the initialized judges, or a no-op Evaluator if
             judge_configuration is None or has no judges
         """
         if not judge_configuration or not judge_configuration.judges:
             return Evaluator.noop()
         judges = self._initialize_judges(
-            judge_configuration.judges, context, default_ai_provider=default_ai_provider
+            judge_configuration.judges, context, default_ai_provider=default_ai_provider,
+            variables=variables,
         )
         return Evaluator(judges, judge_configuration)
 
@@ -604,6 +607,7 @@ def agent_graph(
         self,
         key: str,
         context: Context,
+        default_ai_provider: Optional[str] = None,
     ) -> AgentGraphDefinition:
         """`
         Retrieve an AI agent graph.
@@ -648,7 +652,8 @@ def graph_tracker_factory() -> AIGraphTracker:
         graph_key_value = key
         agent_configs = {
             agent_key: self.__evaluate_agent(
-                agent_key, context, AIAgentConfigDefault.disabled(), graph_key=graph_key_value
+                agent_key, context, AIAgentConfigDefault.disabled(), graph_key=graph_key_value,
+                default_ai_provider=default_ai_provider,
             )
             for agent_key in all_agent_keys
         }
@@ -756,7 +761,7 @@ async def create_agent_graph(
         self._client.track(_TRACK_USAGE_CREATE_AGENT_GRAPH, context, key, 1)
         log.debug(f"Creating managed agent graph for key: {key}")
 
-        graph = self.agent_graph(key, context)
+        graph = self.agent_graph(key, context, default_ai_provider)
         if not graph.enabled:
             return None
 
@@ -916,7 +921,7 @@ def __evaluate_agent(
 
         effective_judge_configuration = judge_configuration or JudgeConfiguration(judges=[])
 
-        evaluator = self._build_evaluator(effective_judge_configuration, context, default_ai_provider)
+        evaluator = self._build_evaluator(effective_judge_configuration, context, default_ai_provider, variables)
 
         return AIAgentConfig(
             key=key,