Merge branch 'main' into feat/add-video-model-capabilities

- Test video-llama and video-llava entries in KNOWN_MODEL_CAPABILITIES
- Test VIDEO capability, VISION, TEXT, STREAMING for video models
- Test VIDEO fallback chain is non-empty with video models
- Test video content routing uses video fallback

MULTIMODAL_BACKLOG.md

@@ -32,4 +32,5 @@
 
 - [x] **Task 1:** Add Gemma 4 entries to `KNOWN_MODEL_CAPABILITIES` and vision fallback chain in `src/infrastructure/models/multimodal.py`. Gemma 4 is a multimodal model supporting vision, text, tools, JSON, and streaming. ✅ PR #1493
 - [x] **Task 3:** Add a `ModelCapability.VIDEO` enum member for future video understanding models. ✅ PR #1494
-- [ ] **Task 4:** Implement `get_model_for_content("video")` routing with appropriate fallback chain.
+- [x] **Task 4:** Implement `get_model_for_content("video")` routing with appropriate fallback chain. ✅ PR #1495
+- [x] **Task 5:** Add VIDEO-capable models to `KNOWN_MODEL_CAPABILITIES` when video understanding models become available in Ollama (e.g., video-llama, video-llava). ✅ (this PR)

src/infrastructure/models/multimodal.py

@@ -192,6 +192,43 @@ KNOWN_MODEL_CAPABILITIES: dict[str, set[ModelCapability]] = {
         ModelCapability.JSON,
         ModelCapability.STREAMING,
     },
+    # Video understanding models (Ollama)
+    "video-llama": {
+        ModelCapability.TEXT,
+        ModelCapability.VISION,
+        ModelCapability.VIDEO,
+        ModelCapability.STREAMING,
+    },
+    "video-llama:7b": {
+        ModelCapability.TEXT,
+        ModelCapability.VISION,
+        ModelCapability.VIDEO,
+        ModelCapability.STREAMING,
+    },
+    "video-llama:13b": {
+        ModelCapability.TEXT,
+        ModelCapability.VISION,
+        ModelCapability.VIDEO,
+        ModelCapability.STREAMING,
+    },
+    "video-llava": {
+        ModelCapability.TEXT,
+        ModelCapability.VISION,
+        ModelCapability.VIDEO,
+        ModelCapability.STREAMING,
+    },
+    "video-llava:7b": {
+        ModelCapability.TEXT,
+        ModelCapability.VISION,
+        ModelCapability.VIDEO,
+        ModelCapability.STREAMING,
+    },
+    "video-llava:13b": {
+        ModelCapability.TEXT,
+        ModelCapability.VISION,
+        ModelCapability.VIDEO,
+        ModelCapability.STREAMING,
+    },
     # Mistral series
     "mistral": {
         ModelCapability.TEXT,
@@ -289,8 +326,11 @@ DEFAULT_FALLBACK_CHAINS: dict[ModelCapability, list[str]] = {
         "moondream:1.8b",  # Tiny vision model (last resort)
     ],
     ModelCapability.VIDEO: [
-        # Video models are not yet available in Ollama
-        # Placeholder for future video understanding models
+        "video-llama:7b",  # Video-LLaMA — video understanding with temporal reasoning
+        "video-llava:7b",  # Video-LLaVA — video understanding via visual-language alignment
+        "video-llama:13b",  # Larger Video-LLaMA (more capable, slower)
+        "video-llava:13b",  # Larger Video-LLaVA
+        "llava:7b",  # Graceful degradation to vision-only model
     ],
 
     ModelCapability.TOOLS: [

tests/infrastructure/test_multimodal.py

@@ -94,6 +94,24 @@ class TestKnownModelCapabilities:
         for name, caps in KNOWN_MODEL_CAPABILITIES.items():
             assert ModelCapability.TEXT in caps, f"{name} should have TEXT"
 
+    def test_video_models_have_video(self):
+        video_names = [
+            "video-llama",
+            "video-llama:7b",
+            "video-llama:13b",
+            "video-llava",
+            "video-llava:7b",
+            "video-llava:13b",
+        ]
+        for name in video_names:
+            assert ModelCapability.VIDEO in KNOWN_MODEL_CAPABILITIES[name], name
+            assert ModelCapability.VISION in KNOWN_MODEL_CAPABILITIES[name], name
+            assert ModelCapability.TEXT in KNOWN_MODEL_CAPABILITIES[name], name
+
+    def test_video_models_have_streaming(self):
+        for name in ["video-llama", "video-llava"]:
+            assert ModelCapability.STREAMING in KNOWN_MODEL_CAPABILITIES[name], name
+
 
 # ---------------------------------------------------------------------------
 # Default fallback chains
@@ -110,6 +128,14 @@ class TestDefaultFallbackChains:
     def test_audio_chain_empty(self):
         assert DEFAULT_FALLBACK_CHAINS[ModelCapability.AUDIO] == []
 
+    def test_video_chain_non_empty(self):
+        assert len(DEFAULT_FALLBACK_CHAINS[ModelCapability.VIDEO]) > 0
+
+    def test_video_chain_has_video_models(self):
+        chain = DEFAULT_FALLBACK_CHAINS[ModelCapability.VIDEO]
+        assert any("video-llama" in m for m in chain)
+        assert any("video-llava" in m for m in chain)
+
 
 # ---------------------------------------------------------------------------
 # Helpers to build a manager without hitting the network
@@ -457,6 +483,12 @@ class TestGetModelForContent:
         model, _ = mgr.get_model_for_content("IMAGE")
         assert model == "llava:7b"
 
+    def test_video_content_uses_video_fallback(self):
+        mgr = _make_manager(["video-llama:7b", "llava:7b"])
+        model, is_fb = mgr.get_model_for_content("video")
+        # Should prefer video-llama from fallback chain
+        assert model == "video-llama:7b"
+
 
 # ---------------------------------------------------------------------------
 # Module-level convenience functions