fix(usage): stamp resolved model id on second-stack usage events

streamHeroku now yields the tier's resolved modelId on its usage event
instead of callers hardcoding kimi-k2-thinking. Removes the now-unused
optionalEnv imports from askDataAgent/analyzeAgent. Addresses final-review
Minor #1 (model mislabel if reasoning tier is ever non-Kimi).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: josers18

lib/inference/analyzeAgent.ts

@@ -25,7 +25,6 @@ import { stripThinkTags } from "@/lib/analyze/sanitize";
 import { extractStructuredFromProse } from "@/lib/analyze/proseToData";
 import { sortRowsByDateLikeColumn } from "@/lib/analyze/sortByDate";
 import { log } from "@/lib/log";
-import { optionalEnv } from "@/lib/utils";
 
 /**
  * Tools Kimi is allowed to see on the Analyze surface. Doc-grounded
@@ -260,7 +259,7 @@ export async function* runAnalyzeAgent(
             inputTokens: ev.inputTokens,
             outputTokens: ev.outputTokens,
             exact: ev.exact,
-            model: optionalEnv("HEROKU_INFERENCE_ONYX_MODEL_ID") || "kimi-k2-thinking",
+            model: ev.model,
           };
         }
       }

lib/inference/askDataAgent.ts

@@ -28,7 +28,6 @@ import type {
   FirstPartyDcToolDef,
 } from "@/lib/mcp/firstPartyDataCloud";
 import { stripThinkTags } from "@/lib/analyze/sanitize";
-import { optionalEnv } from "@/lib/utils";
 
 // Soft budget: if the agent hasn't answered after N tool-call iterations
 // we bail. Heroku router already stalls long requests — this keeps us well
@@ -223,7 +222,7 @@ export async function* runAskDataAgent(
             inputTokens: ev.inputTokens,
             outputTokens: ev.outputTokens,
             exact: ev.exact,
-            model: optionalEnv("HEROKU_INFERENCE_ONYX_MODEL_ID") || "kimi-k2-thinking",
+            model: ev.model,
           };
         }
       }

lib/inference/heroku.ts

@@ -51,6 +51,7 @@ export type HerokuInferenceEvent =
   | { type: "done"; stopReason: string | null }
   | {
       type: "usage";
+      model: string;
       inputTokens: number;
       outputTokens: number;
       exact: boolean;
@@ -297,15 +298,20 @@ export async function* streamHeroku(
     };
   }
 
+  // why: stamp the usage event with the tier's RESOLVED model id (from
+  // clientFor) rather than letting callers guess — otherwise a tier served
+  // by a non-Kimi backend would be mislabeled in the token-spend panel.
   yield usageAcc.exact
     ? {
         type: "usage" as const,
+        model: modelId,
         inputTokens: usageAcc.inputTokens,
         outputTokens: usageAcc.outputTokens,
         exact: true,
       }
     : {
         type: "usage" as const,
+        model: modelId,
         inputTokens: estimateTokens(
           messages
             .map((m) =>
@@ -320,9 +326,6 @@ export async function* streamHeroku(
       };
 
   yield { type: "done", stopReason };
-  // Avoid "modelId was defined but unused" false-negative — it's surfaced
-  // on the caller via inferHeroku's result, not through the stream.
-  void modelId;
 }
 
 function safeParseJson(s: string): unknown {