fix(notifications): inject llm_backend into build_pipeline

build_pipeline previously constructed its own LLMBackend from
config.llm, which:
  - duplicated main.py's already-running backend instance
  - failed to inherit env-loaded LLM_API_KEY when called from
    short-lived scripts (eyeball checks, tests), forcing fallback
  - prevented pipeline components from sharing the live backend

build_pipeline and build_pipeline_components now require an
llm_backend parameter. main.py passes the same instance it
constructed for its primary responder. Tests pass mocks. The
digest accumulator now uses the live, authenticated backend.

Added test_build_pipeline_uses_provided_backend to lock in the
injection contract.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

meshai/notifications/pipeline/__init__.py

@@ -12,7 +12,7 @@ Phase 2.4:
 
 Usage:
     from meshai.notifications.pipeline import build_pipeline, start_pipeline, stop_pipeline
-    bus = build_pipeline(config)
+    bus = build_pipeline(config, llm_backend)  # llm_backend from main.py
     bus.emit(event)
 
     # Async lifecycle
@@ -35,40 +35,20 @@ from meshai.notifications.pipeline.digest import DigestAccumulator, Digest
 from meshai.notifications.pipeline.scheduler import DigestScheduler
 
 
-def _create_llm_backend(config):
+def build_pipeline(config, llm_backend) -> EventBus:
-    """Create an LLM backend from config, or return None if unavailable."""
-    try:
-        from meshai.backends import OpenAIBackend, AnthropicBackend, GoogleBackend
-
-        api_key = config.resolve_api_key()
-        if not api_key:
-            return None
-
-        backend_name = config.llm.backend.lower()
-        # Use minimal memory settings for digest summaries
-        if backend_name == "openai":
-            return OpenAIBackend(config.llm, api_key, 0, 0)
-        elif backend_name == "anthropic":
-            return AnthropicBackend(config.llm, api_key, 0, 0)
-        elif backend_name == "google":
-            return GoogleBackend(config.llm, api_key, 0, 0)
-        else:
-            return OpenAIBackend(config.llm, api_key, 0, 0)
-    except Exception:
-        return None
-
-
-def build_pipeline(config) -> EventBus:
     """Build the pipeline and return the EventBus.
 
+    Args:
+        config: Full Config object.
+        llm_backend: An already-constructed LLMBackend instance
+            (from main.py or a test). Pipeline components share
+            this single instance. May be None for fallback behavior.
+
     Components are stashed on bus._pipeline_components for lifecycle use.
     """
     bus = EventBus()
     dispatcher = Dispatcher(config, create_channel)
 
-    # Build LLM backend for digest summarization
-    llm_backend = _create_llm_backend(config)
-
     # Build include_toggles from config
     digest_cfg = getattr(config.notifications, "digest", None)
     include_toggles = None
@@ -116,17 +96,21 @@ def build_pipeline(config) -> EventBus:
     return bus
 
 
-def build_pipeline_components(config) -> tuple:
+def build_pipeline_components(config, llm_backend) -> tuple:
     """Like build_pipeline, but returns all components for tests.
 
-    Returns (bus, inhibitor, grouper, toggle_filter, dispatcher, accumulator).
+    Args:
+        config: Full Config object.
+        llm_backend: An already-constructed LLMBackend instance
+            (from main.py or a test). Pipeline components share
+            this single instance. May be None for fallback behavior.
+
+    Returns:
+        (bus, inhibitor, grouper, toggle_filter, dispatcher, accumulator).
     """
     bus = EventBus()
     dispatcher = Dispatcher(config, create_channel)
 
-    # Build LLM backend for digest summarization
-    llm_backend = _create_llm_backend(config)
-
     # Build include_toggles from config
     digest_cfg = getattr(config.notifications, "digest", None)
     include_toggles = None

tests/test_pipeline_digest.py

@@ -50,6 +50,13 @@ class FailingLLMBackend:
         raise RuntimeError("LLM unavailable")
 
 
+def _make_mock_backend():
+    """Create a standard mock LLM backend for tests."""
+    mock = MagicMock()
+    mock.generate = AsyncMock(return_value="stub summary")
+    return mock
+
+
 # ============================================================
 # ACCUMULATOR EVENT LOGGING TESTS
 # ============================================================
@@ -478,7 +485,7 @@ def test_pipeline_routes_event_to_accumulator():
     """Events via bus.emit end up in DigestAccumulator."""
     config = Config()
     bus, inhibitor, grouper, toggle_filter, dispatcher, accumulator = \
-        build_pipeline_components(config)
+        build_pipeline_components(config, _make_mock_backend())
 
     event = make_event(
         source="test",
@@ -499,7 +506,7 @@ def test_pipeline_routes_immediate_to_both():
     """Immediate events go to both dispatcher and accumulator in Phase 2.4."""
     config = Config()
     bus, inhibitor, grouper, toggle_filter, dispatcher, accumulator = \
-        build_pipeline_components(config)
+        build_pipeline_components(config, _make_mock_backend())
 
     event = make_event(
         source="test",

tests/test_pipeline_toggle_filter.py

@@ -1,6 +1,7 @@
 """Tests for ToggleFilter (Phase 2.4)."""
 
 import pytest
+from unittest.mock import MagicMock, AsyncMock
 
 from meshai.notifications.events import make_event
 from meshai.notifications.pipeline.toggle_filter import ToggleFilter
@@ -101,16 +102,31 @@ class TestToggleFilterPipelineWiring:
 
     def test_toggle_filter_pipeline_drops_disabled_toggle(self):
         """Events for disabled toggles don't reach dispatcher or accumulator."""
-        # Create config with only weather enabled
         config = Config()
-        # We'll check by using build_pipeline_components and inspecting
+
-        # In Phase 2.4, build_pipeline_components returns toggle_filter
+        # Pass mock LLM backend
+        mock_backend = MagicMock()
+        mock_backend.generate = AsyncMock(return_value="stub summary")
 
         # Note: without toggles.enabled set, filter is a no-op
         # This test verifies the wiring is correct
-        bus, inhibitor, grouper, toggle_filter, dispatcher, accumulator = \
+        bus, inhibitor, grouper, toggle_filter, dispatcher, accumulator =             build_pipeline_components(config, mock_backend)
-            build_pipeline_components(config)
 
         # Verify toggle_filter is in the chain
         assert toggle_filter is not None
         assert hasattr(toggle_filter, 'handle')
+
+    def test_build_pipeline_uses_provided_backend(self):
+        """build_pipeline_components uses the provided llm_backend."""
+        config = Config()
+
+        # Sentinel backend with unique attribute
+        sentinel = MagicMock()
+        sentinel.unique_marker = "I_AM_THE_SENTINEL"
+        sentinel.generate = AsyncMock(return_value="sentinel summary")
+
+        bus, inhibitor, grouper, toggle_filter, dispatcher, accumulator =             build_pipeline_components(config, sentinel)
+
+        # Accumulator should have the exact sentinel instance
+        assert accumulator._llm is sentinel
+        assert accumulator._llm.unique_marker == "I_AM_THE_SENTINEL"