Remove corpus counts from chat chrome

apps/ai-service/routers/rag.py

@@ -40,6 +40,25 @@ class CitationResponse(BaseModel):
     evidence_text: str = ""
 
 
+class AnswerClaimResponse(BaseModel):
+    text: str
+    source_ids: list[str]
+
+
+class AnswerBlockResponse(BaseModel):
+    title: str
+    kind: str
+    claims: list[AnswerClaimResponse]
+
+
+class AnswerPlanResponse(BaseModel):
+    verbosity: str
+    layout: str
+    reasoning_mode: str
+    show_heading: bool
+    needs_warning: bool
+
+
 class RagQueryResponse(BaseModel):
     trace_id: str
     decision: str
@@ -57,6 +76,9 @@ class RagQueryResponse(BaseModel):
     # clarify path today; other clarify sources (no_drug, dosing_calc's
     # needs_clarify) leave this empty rather than fabricate options.
     quick_replies: list[str] = []
+    blocks: list[AnswerBlockResponse] = []
+    answer_mode: str = "concise"
+    answer_plan: AnswerPlanResponse | None = None
 
 
 def _answer_service(request: Request) -> GroundedAnswerService:
@@ -110,6 +132,32 @@ def _map_citations(items) -> list[CitationResponse]:
     ]
 
 
+def _map_blocks(items) -> list[AnswerBlockResponse]:
+    return [
+        AnswerBlockResponse(
+            title=item.title,
+            kind=item.kind,
+            claims=[
+                AnswerClaimResponse(text=claim.text, source_ids=list(claim.source_ids))
+                for claim in item.claims
+            ],
+        )
+        for item in items
+    ]
+
+
+def _map_plan(item) -> AnswerPlanResponse | None:
+    if item is None:
+        return None
+    return AnswerPlanResponse(
+        verbosity=item.verbosity,
+        layout=item.layout,
+        reasoning_mode=item.reasoning_mode,
+        show_heading=item.show_heading,
+        needs_warning=item.needs_warning,
+    )
+
+
 @router.post("/query", response_model=RagQueryResponse)
 def query_rag(
     payload: RagQueryRequest,
@@ -144,6 +192,9 @@ def query_rag(
         citations = _map_citations(reply.citations)
         generated = reply.generated
         quick_replies = list(reply.quick_replies)
+        blocks = _map_blocks(reply.blocks)
+        answer_mode = reply.answer_mode
+        answer_plan = _map_plan(reply.plan)
     else:
         # No generator configured (ANSWER_PROVIDER=disabled): there is no LLM
         # to understand a turn with, so this is retrieval-only, single-turn,
@@ -156,6 +207,9 @@ def query_rag(
             citations = []
             generated = False
             quick_replies = list(grounded.quick_replies)
+            blocks = []
+            answer_mode = "concise"
+            answer_plan = None
         else:
             decision = grounded.result.decision.value
             reason = grounded.result.reason
@@ -164,6 +218,9 @@ def query_rag(
             citations = _map_citations(grounded.citations)
             generated = grounded.generated
             quick_replies = []
+            blocks = _map_blocks(grounded.blocks)
+            answer_mode = grounded.answer_mode
+            answer_plan = _map_plan(grounded.plan)
 
     # Trace persistence is fail-open (F-09): an already-computed, safe answer
     # must reach the caller even if Postgres is unreachable. `save()` opens a
@@ -197,4 +254,7 @@ def query_rag(
         citations=citations,
         generated=generated,
         quick_replies=quick_replies,
+        blocks=blocks,
+        answer_mode=answer_mode,
+        answer_plan=answer_plan,
     )