diff --git a/assistant/src/config/schemas/llm.ts b/assistant/src/config/schemas/llm.ts
index 4c4800be790..8cfaab22f1a 100644
--- a/assistant/src/config/schemas/llm.ts
+++ b/assistant/src/config/schemas/llm.ts
@@ -65,6 +65,8 @@ export const LLMCallSiteEnum = z.enum([
   "styleAnalyzer",
   "inviteInstructionGenerator",
   "skillCategoryInference",
+  "meetConsentMonitor",
+  "meetChatOpportunity",
 ]);
 export type LLMCallSite = z.infer<typeof LLMCallSiteEnum>;
 
diff --git a/clients/macos/vellum-assistant/Features/Settings/CallSiteOverride.swift b/clients/macos/vellum-assistant/Features/Settings/CallSiteOverride.swift
index 7fc7543e16e..ce832fb1af7 100644
--- a/clients/macos/vellum-assistant/Features/Settings/CallSiteOverride.swift
+++ b/clients/macos/vellum-assistant/Features/Settings/CallSiteOverride.swift
@@ -14,6 +14,7 @@ public enum CallSiteDomain: String, CaseIterable, Identifiable, Hashable {
     case notifications
     case voice
     case utility
+    case skills
 
     public var id: String { rawValue }
 
@@ -28,6 +29,7 @@ public enum CallSiteDomain: String, CaseIterable, Identifiable, Hashable {
         case .notifications: return "Notifications"
         case .voice: return "Voice"
         case .utility: return "Utility"
+        case .skills: return "Skills"
         }
     }
 
@@ -42,6 +44,7 @@ public enum CallSiteDomain: String, CaseIterable, Identifiable, Hashable {
         case .notifications: return 4
         case .voice: return 5
         case .utility: return 6
+        case .skills: return 7
         }
     }
 }
@@ -140,6 +143,9 @@ public enum CallSiteCatalog {
         CallSiteOverride(id: "styleAnalyzer", displayName: "Style analyzer", domain: .utility),
         CallSiteOverride(id: "inviteInstructionGenerator", displayName: "Invite instruction generator", domain: .utility),
         CallSiteOverride(id: "skillCategoryInference", displayName: "Skill category inference", domain: .utility),
+        // Skills
+        CallSiteOverride(id: "meetConsentMonitor", displayName: "Meet · Consent monitor", domain: .skills),
+        CallSiteOverride(id: "meetChatOpportunity", displayName: "Meet · Chat opportunity", domain: .skills),
     ]
 
     /// Lookup table from call-site ID to its catalog entry. Constructed
diff --git a/skills/meet-join/daemon/consent-monitor.ts b/skills/meet-join/daemon/consent-monitor.ts
index 5689aba690b..4da97335ab9 100644
--- a/skills/meet-join/daemon/consent-monitor.ts
+++ b/skills/meet-join/daemon/consent-monitor.ts
@@ -127,8 +127,9 @@ export interface MeetConsentMonitorDeps {
   config: MeetConsentMonitorConfig;
   /**
    * Ask the LLM for an objection verdict. Defaults to a wrapper around the
-   * repo-wide provider abstraction using {@link CONSENT_LLM_MAX_TOKENS} and
-   * `modelIntent: "latency-optimized"`. Tests inject scripted responses.
+   * repo-wide provider abstraction using {@link CONSENT_LLM_MAX_TOKENS}
+   * under the `meetConsentMonitor` call site. Tests inject scripted
+   * responses.
    */
   llmAsk?: ObjectionLLMAsk;
   /** Override the dispatcher subscribe (tests). */
@@ -693,7 +694,7 @@ const OBJECTION_TOOL: ToolDefinition = {
 
 /**
  * Default {@link ObjectionLLMAsk} — routes through the repo-wide provider
- * abstraction with `modelIntent: "latency-optimized"`, times out at
+ * abstraction under the `meetConsentMonitor` call site, times out at
  * {@link CONSENT_LLM_TIMEOUT_MS}, and extracts the tool-use input as the
  * structured verdict.
  *
@@ -701,7 +702,8 @@ const OBJECTION_TOOL: ToolDefinition = {
  * real provider.
  */
 async function defaultLLMAsk(prompt: string): Promise<ObjectionDecision> {
-  const provider: Provider | null = await getConfiguredProvider();
+  const provider: Provider | null =
+    await getConfiguredProvider("meetConsentMonitor");
   if (!provider) {
     // No provider available — conservatively assume no objection so the
     // monitor doesn't interrupt a meeting based on missing infra.
@@ -716,7 +718,7 @@ async function defaultLLMAsk(prompt: string): Promise<ObjectionDecision> {
       "You are a strict JSON classifier. Only respond via the report_objection tool.",
       {
         config: {
-          modelIntent: "latency-optimized",
+          callSite: "meetConsentMonitor",
           max_tokens: CONSENT_LLM_MAX_TOKENS,
           tool_choice: { type: "tool" as const, name: OBJECTION_TOOL.name },
         },
diff --git a/skills/meet-join/daemon/session-manager.ts b/skills/meet-join/daemon/session-manager.ts
index 3cc831558a8..07b47b779ba 100644
--- a/skills/meet-join/daemon/session-manager.ts
+++ b/skills/meet-join/daemon/session-manager.ts
@@ -505,9 +505,9 @@ export interface MeetSessionManagerDeps {
   /**
    * Override the chat-opportunity-detector factory. Default constructs a
    * {@link MeetChatOpportunityDetector} with a Tier 2 LLM callback that
-   * routes through the repo-wide provider abstraction at
-   * `modelIntent: "latency-optimized"`. Tests can inject a fake to
-   * observe start/dispose/stats without spinning up the LLM path.
+   * routes through the repo-wide provider abstraction under the
+   * `meetChatOpportunity` call site. Tests can inject a fake to observe
+   * start/dispose/stats without spinning up the LLM path.
    *
    * Only consulted when `services.meet.proactiveChat.enabled === true`.
    */
@@ -1689,10 +1689,9 @@ const CHAT_OPPORTUNITY_TOOL: ToolDefinition = {
 
 /**
  * Default Tier 2 chat-opportunity LLM callback. Routes through the
- * repo-wide provider abstraction at
- * `modelIntent: "latency-optimized"` — keeping the proactive-chat path
- * on the same latency tier the consent monitor uses so both background
- * loops share tuning. Times out at
+ * repo-wide provider abstraction under the `meetChatOpportunity` call
+ * site, keeping the proactive-chat path on its own configurable lane
+ * alongside the consent monitor. Times out at
  * {@link CHAT_OPPORTUNITY_LLM_TIMEOUT_MS} and extracts the tool-use
  * input as the structured verdict.
  *
@@ -1703,7 +1702,8 @@ const CHAT_OPPORTUNITY_TOOL: ToolDefinition = {
 async function defaultCallDetectorLLM(
   prompt: string,
 ): Promise<ChatOpportunityDecision> {
-  const provider: Provider | null = await getConfiguredProvider();
+  const provider: Provider | null =
+    await getConfiguredProvider("meetChatOpportunity");
   if (!provider) {
     return { shouldRespond: false, reason: "" };
   }
@@ -1716,7 +1716,7 @@ async function defaultCallDetectorLLM(
       "You are a strict JSON classifier. Only respond via the report_chat_opportunity tool.",
       {
         config: {
-          modelIntent: "latency-optimized",
+          callSite: "meetChatOpportunity",
           max_tokens: CHAT_OPPORTUNITY_LLM_MAX_TOKENS,
           tool_choice: {
             type: "tool" as const,