feat: Add semantic caching (both generic and scoped) to the demo workflow (optional)

.env.example

@@ -24,7 +24,7 @@ CTX_REDIS_INSTANCE_ID=
 DEMO_USER_ID=
 DEMO_USER_NAME=
 DEMO_USER_EMAIL=
-# Available domains: reddash, electrohub, healthcare
+# Available domains: reddash, electrohub, healthcare, airline-support, finance-researcher
 DEMO_DOMAIN=reddash
 
 # ═══════════════════════════════════════════════════════

README.md

@@ -21,9 +21,9 @@ The repo currently includes built-in demo domains for:
 
 - `reddash` — food-delivery support
 - `electrohub` — electronics retail and order support
+- `airline-support` — passenger support across disruptions, reassignment, traveller profile lookup, and scoped semantic caching
 - `finance-researcher` — ShiftIQ watchlist research across filings, metrics, prices, and live updates
 - `healthcare` — RedHealthConnect patient success portal (appointments, referrals, providers)
-- `airline-support` — synthetic airline support across disruptions, reassignment, and traveller profile lookup
 
 **Two modes, same UI:**
 
@@ -126,15 +126,17 @@ Open http://localhost:3040 and try:
 - In `electrohub`:
   - *"Show me my recent ElectroHub orders."*
   - *"Can I pick that up at my local store?"*
+- In `airline-support`:
+  - *"My flight was disrupted. What happened?"*
+  - *"Can I still check in for the new flight?"*
+  - *"What help do I usually get after a cancellation?"*
+  - *"What is the status of ZX018 today?"*
 - In `finance-researcher` / ShiftIQ:
   - *"Walk me through Oracle's latest quarter using both the filing and the structured metrics."*
   - *"What's new in my watchlist this week?"*
 - In `healthcare`:
   - *"Do I have any upcoming appointments?"*
   - *"Find me a cardiologist accepting new patients?"*
-- In `airline-support`:
-  - *"My flight was disrupted. What happened?"*
-  - *"Can I still check in for the new flight?"*
 
 ---
 
@@ -214,9 +216,9 @@ See:
 
 - [`domains/reddash/docs/demo_paths.md`](domains/reddash/docs/demo_paths.md)
 - [`domains/electrohub/docs/demo_paths.md`](domains/electrohub/docs/demo_paths.md)
+- [`domains/airline-support/docs/demo_paths.md`](domains/airline-support/docs/demo_paths.md)
 - [`domains/finance-researcher/docs/demo_paths.md`](domains/finance-researcher/docs/demo_paths.md)
 - [`domains/healthcare/docs/demo_paths.md`](domains/healthcare/docs/demo_paths.md)
-- [`domains/airline-support/docs/demo_paths.md`](domains/airline-support/docs/demo_paths.md)
 
 Reddash includes four scripted conversation flows:
 
@@ -227,19 +229,20 @@ Reddash includes four scripted conversation flows:
 
 > **Tip:** After each path, toggle to Simple RAG mode and ask the same question to see the contrast.
 
+Airline support includes flagship and semantic-cache paths for:
+
+1. **Disruption recovery** — explain a cancelled flight, the updated itinerary, and next steps
+2. **Post-rebooking serviceability** — answer check-in, baggage, and terminal questions against the reassigned trip
+3. **Traveller profile snapshot** — backup-only path for read-only identity and profile grounding
+4. **Scoped semantic caching** — demonstrate public flight-status reuse, cohort-scoped passenger service guidance, and personalized support that intentionally stays out of the cache
+
 ShiftIQ includes flagship paths for:
 
 1. **Cross-company narrative comparison** — compare the latest filings and research chunks across peers
 2. **Metric-plus-document reasoning** — explain a quarter using both structured metrics and source documents
 3. **Peer trend analysis** — compare price and fundamentals trends, including RedisTimeSeries-backed queries
 4. **Live watchlist updates** — explain what changed this week using normalized coverage events and Redis Streams
 
-Airline support includes flagship paths for:
-
-1. **Disruption recovery** — explain a cancelled flight, the updated itinerary, and next steps
-2. **Post-rebooking serviceability** — answer check-in, baggage, and terminal questions against the reassigned trip
-3. **Traveller profile snapshot** — backup-only path for read-only identity and profile grounding
-
 ## Presentations
 
 Keep domain-specific presentations with the domain itself:
@@ -290,9 +293,9 @@ context-engine-demos/
 ├── domains/
 │   ├── reddash/             # Delivery-support reference domain
 │   ├── electrohub/          # Electronics retail reference domain
+│   ├── airline-support/     # Passenger support + semantic cache reference domain
 │   ├── finance-researcher/  # ShiftIQ watchlist research domain
-│   ├── healthcare/          # Patient success portal domain
-│   └── airline-support/     # Airline disruption and trip support domain
+│   └── healthcare/          # Patient success portal domain
 │       ├── domain.py        # DOMAIN export implementing the contract
 │       ├── schema.py        # Entity definitions
 │       ├── prompt.py        # Domain prompt/playbooks

backend/app/contracts.py

@@ -14,4 +14,4 @@ class ChatRequest(BaseModel):
     messages: list[ChatMessage]
     mode: Literal["context_surfaces", "simple_rag"] = "context_surfaces"
     thread_id: str | None = None
-
+    demo_user_id: str | None = None

backend/app/core/domain_contract.py

@@ -87,6 +87,26 @@ class IdentityConfig(BaseModel):
     description: str
 
 
+class SemanticCacheConfig(BaseModel):
+    enabled: bool = False
+    cache_name: str = ""
+    distance_threshold: float = 0.12
+    ttl_seconds: int = 1800
+
+
+class InternalToolAccessControl(BaseModel):
+    access_control_enabled: bool = False
+    touched_entity_classes: list[str] = Field(default_factory=list)
+    access_class_override: str | None = None
+
+
+class DemoUserOption(BaseModel):
+    id: str
+    label: str
+    subtitle: str | None = None
+    cache_group_id: str | None = None
+
+
 class DomainManifest(BaseModel):
     id: str
     version: str = "1"
@@ -98,12 +118,14 @@ class DomainManifest(BaseModel):
     namespace: NamespaceConfig
     rag: RagConfig
     identity: IdentityConfig
+    semantic_cache: SemanticCacheConfig = Field(default_factory=SemanticCacheConfig)
 
 
 class InternalToolDefinition(BaseModel):
     name: str
     description: str
     input_schema: dict[str, Any] = Field(default_factory=lambda: {"type": "object", "properties": {}, "required": []})
+    access_control: InternalToolAccessControl = Field(default_factory=InternalToolAccessControl)
 
 
 class GeneratedDataset(BaseModel):

backend/app/langgraph_agent.py

@@ -362,6 +362,10 @@ async def create_checkpointer(settings: Settings) -> AsyncRedisSaver:
     domain = get_active_domain(settings)
     checkpointer = AsyncRedisSaver(
         redis_url=redis_url,
+        connection_args={
+            "max_connections": settings.redis_max_connections,
+            "health_check_interval": 30,
+        },
         checkpoint_prefix=domain.manifest.namespace.checkpoint_prefix,
         checkpoint_write_prefix=domain.manifest.namespace.checkpoint_write_prefix,
     )