Gate combined-chat capability vision on the driving model

haiku_rag_slim/haiku/rag/capabilities/analysis.py

@@ -13,7 +13,7 @@ from haiku.rag.capabilities._base import (
     RAGCapabilityBase,
     resolve_db_path,
 )
-from haiku.rag.config.models import AppConfig
+from haiku.rag.config.models import AppConfig, ModelConfig
 from haiku.rag.sandbox import AnalysisContext, Sandbox
 from haiku.rag.store.models.chunk import SearchResult
 from haiku.rag.store.models.citation import Citation
@@ -129,8 +129,14 @@ def create_capability(
     *,
     defer_loading: bool = True,
     request_limit: int | None = 30,
+    model: ModelConfig | None = None,
 ) -> AnalysisCapability:
-    """Create a native Pydantic AI analysis capability."""
+    """Create a native Pydantic AI analysis capability.
+
+    ``model`` sets the capability's image-attachment gate and should be the
+    model the hosting agent actually runs. Defaults to ``config.analysis.model``
+    (falling back to ``config.qa.model``).
+    """
     if config is None:
         from haiku.rag.config import get_config
 
@@ -141,7 +147,7 @@ def create_capability(
         state_type=AnalysisState,
         state_namespace=STATE_NAMESPACE,
         instruction_text=instructions(),
-        model=config.analysis.model or config.qa.model,
+        model=model or config.analysis.model or config.qa.model,
         tool_names=_TOOL_NAMES,
         request_limit=request_limit,
         id=_CAPABILITY_ID,

haiku_rag_slim/haiku/rag/capabilities/rag.py

@@ -12,7 +12,7 @@ from haiku.rag.capabilities._base import (
     RAGCapabilityBase,
     resolve_db_path,
 )
-from haiku.rag.config.models import AppConfig
+from haiku.rag.config.models import AppConfig, ModelConfig
 from haiku.rag.store.models.chunk import SearchResult
 from haiku.rag.store.models.citation import Citation
 
@@ -71,8 +71,13 @@ def create_capability(
     *,
     defer_loading: bool = True,
     request_limit: int | None = 20,
+    model: ModelConfig | None = None,
 ) -> RAGCapability:
-    """Create a native Pydantic AI RAG capability."""
+    """Create a native Pydantic AI RAG capability.
+
+    ``model`` sets the capability's image-attachment gate and should be the
+    model the hosting agent actually runs. Defaults to ``config.qa.model``.
+    """
     if config is None:
         from haiku.rag.config import get_config
 
@@ -83,7 +88,7 @@ def create_capability(
         state_type=RAGState,
         state_namespace=STATE_NAMESPACE,
         instruction_text=instructions(),
-        model=config.qa.model,
+        model=model or config.qa.model,
         tool_names=_TOOL_NAMES,
         request_limit=request_limit,
         id=_CAPABILITY_ID,

haiku_rag_slim/haiku/rag/chat/__init__.py

@@ -38,12 +38,24 @@ def run_chat(
     capability_list = []
     defer_loading = len(enabled) > 1
 
+    # One agent drives every attached capability, so a capability's
+    # image-attachment gate must track that single model: analysis.model only
+    # when analysis runs alone, otherwise qa.model. Passing it to every
+    # capability keeps their vision flag aligned with the model actually running.
+    if "rag" not in enabled and "analysis" in enabled:
+        driving_model = config.analysis.model or config.qa.model
+    else:
+        driving_model = config.qa.model
+
     if "rag" in enabled:
         from haiku.rag.capabilities.rag import create_capability
 
         capability_list.append(
             create_capability(
-                db_path=db_path, config=config, defer_loading=defer_loading
+                db_path=db_path,
+                config=config,
+                defer_loading=defer_loading,
+                model=driving_model,
             )
         )
 
@@ -52,18 +64,13 @@ def run_chat(
 
         capability_list.append(
             create_capability(
-                db_path=db_path, config=config, defer_loading=defer_loading
+                db_path=db_path,
+                config=config,
+                defer_loading=defer_loading,
+                model=driving_model,
             )
         )
 
-    # Drive with the analysis model when analysis is the only capability, so the
-    # running model matches the one the analysis capability configures (including
-    # its vision flag). RAG runs on the QA model.
-    if "rag" not in enabled and "analysis" in enabled:
-        driving_model = config.analysis.model or config.qa.model
-    else:
-        driving_model = config.qa.model
-
     app = ChatApp(
         db_path,
         capabilities=capability_list,

tests/chat/test_chat_app.py

@@ -54,10 +54,11 @@ def test_run_chat_defers_multiple_capabilities(temp_db_path: Path):
         (["rag", "analysis"], "qa-model"),
     ],
 )
-def test_run_chat_drives_analysis_only_with_analysis_model(
+def test_run_chat_gates_capability_vision_on_driving_model(
     temp_db_path: Path, enabled, expected_model
 ):
-    """Analysis-only chat runs on analysis.model; otherwise on qa.model."""
+    """Analysis-only chat runs on analysis.model; otherwise on qa.model. Every
+    attached capability's vision gate (its ``model``) tracks that one model."""
     from haiku.rag.config.models import AppConfig, ModelConfig
 
     config = AppConfig()
@@ -70,7 +71,7 @@ def test_run_chat_drives_analysis_only_with_analysis_model(
         return "resolved-model"
 
     with (
-        patch("haiku.rag.chat.app.ChatApp"),
+        patch("haiku.rag.chat.app.ChatApp") as mock_app,
         patch("haiku.rag.config.get_config", return_value=config),
         patch("haiku.rag.utils.get_model", side_effect=fake_get_model),
     ):
@@ -79,6 +80,8 @@ def test_run_chat_drives_analysis_only_with_analysis_model(
         run_chat(db_path=temp_db_path, capabilities=enabled)
 
     assert captured["name"] == expected_model
+    attached = mock_app.call_args.kwargs["capabilities"]
+    assert {capability.model.name for capability in attached} == {expected_model}
 
 
 def _make_mock_client():