@tonyclaw/agent-inspector 2.0.28 → 2.0.29

package/src/routes/api/logs.$id.replay.ts

@@ -21,6 +21,7 @@ import {
   selectUpstreamBase,
   setUpstreamHost,
 } from "../../proxy/upstream";
+import { computeTokensPerSecond } from "../../proxy/streamTiming";
 
 type ReplayRequest = {
   modifiedBody: string;
@@ -38,6 +39,9 @@ type ReplayResponse = {
   inputTokens?: number;
   outputTokens?: number;
   elapsedMs?: number;
+  firstChunkMs?: number | null;
+  totalStreamMs?: number | null;
+  tokensPerSecond?: number | null;
   streaming?: boolean;
   replayLogId?: number;
 };
@@ -135,7 +139,6 @@ export const Route = createFileRoute("/api/logs/$id/replay")({
           } satisfies ReplayResponse);
         }
 
-        const elapsedMs = Date.now() - startTime;
         const isStream =
           upstreamRes.headers.get(HEADER_CONTENT_TYPE)?.includes(CONTENT_TYPE_EVENT_STREAM) ??
           false;
@@ -145,11 +148,13 @@ export const Route = createFileRoute("/api/logs/$id/replay")({
           const chunks: string[] = [];
           const decoder = new TextDecoder();
           const reader = upstreamRes.body?.getReader();
+          let firstChunkMs: number | null = null;
           if (reader) {
             try {
               while (true) {
                 const { done, value } = await reader.read();
                 if (done) break;
+                if (firstChunkMs === null) firstChunkMs = Date.now() - startTime;
                 chunks.push(decoder.decode(value, { stream: true }));
               }
             } catch {
@@ -157,11 +162,19 @@ export const Route = createFileRoute("/api/logs/$id/replay")({
             }
           }
           const fullResponse = chunks.join("");
+          const elapsedMs = Date.now() - startTime;
+          const totalStreamMs = elapsedMs;
 
           // Create a mock log for streaming extraction
           const mockLog: typeof log = { ...log };
           const responseText = formatHandler.extractStream(fullResponse, mockLog, model, false);
           const tokens = formatHandler.extractTokens(responseText);
+          const outputTokens = tokens.outputTokens ?? null;
+          const tokensPerSecond = computeTokensPerSecond({
+            outputTokens,
+            firstChunkMs,
+            totalStreamMs,
+          });
           const savedReplayLog = await addTestLogEntry({
             timestamp: new Date().toISOString(),
             method: log.method,
@@ -172,10 +185,13 @@ export const Route = createFileRoute("/api/logs/$id/replay")({
             responseStatus: upstreamRes.status,
             responseText,
             inputTokens: tokens.inputTokens ?? null,
-            outputTokens: tokens.outputTokens ?? null,
+            outputTokens,
             cacheCreationInputTokens: tokens.cacheCreationInputTokens ?? null,
             cacheReadInputTokens: tokens.cacheReadInputTokens ?? null,
             elapsedMs,
+            firstChunkMs,
+            totalStreamMs,
+            tokensPerSecond,
             streaming: true,
             userAgent: log.userAgent,
             origin: log.origin,
@@ -201,11 +217,15 @@ export const Route = createFileRoute("/api/logs/$id/replay")({
             inputTokens: tokens.inputTokens ?? undefined,
             outputTokens: tokens.outputTokens ?? undefined,
             elapsedMs,
+            firstChunkMs,
+            totalStreamMs,
+            tokensPerSecond,
             streaming: true,
             replayLogId: savedReplayLog.id,
           } satisfies ReplayResponse);
         } else {
           const responseText = await upstreamRes.text();
+          const elapsedMs = Date.now() - startTime;
           const tokens = formatHandler.extractTokens(responseText);
           const savedReplayLog = await addTestLogEntry({
             timestamp: new Date().toISOString(),
@@ -221,6 +241,9 @@ export const Route = createFileRoute("/api/logs/$id/replay")({
             cacheCreationInputTokens: tokens.cacheCreationInputTokens ?? null,
             cacheReadInputTokens: tokens.cacheReadInputTokens ?? null,
             elapsedMs,
+            firstChunkMs: null,
+            totalStreamMs: null,
+            tokensPerSecond: null,
             streaming: false,
             userAgent: log.userAgent,
             origin: log.origin,

package/src/routes/api/providers.$providerId.test.log.ts

@@ -8,6 +8,7 @@ import {
   type ProviderTestResult as TestResult,
   type ProviderTestState,
 } from "../../lib/providerTestContract";
+import { computeTokensPerSecond, firstChunkMsFromChunks } from "../../proxy/streamTiming";
 
 function hasSuccessField(result: ProviderTestState): result is TestResult {
   return Object.prototype.hasOwnProperty.call(result, "success");
@@ -29,6 +30,25 @@ function getResultHeaders(result: TestResult): Record<string, string> {
   return result.requestHeaders ?? result.debug?.requestHeaders ?? {};
 }
 
+function getResultTotalStreamMs(result: TestResult): number | null {
+  return result.totalStreamMs ?? (result.streaming === true ? (result.latencyMs ?? null) : null);
+}
+
+function getResultFirstChunkMs(result: TestResult): number | null {
+  return result.firstChunkMs ?? firstChunkMsFromChunks(result.streamingChunks?.chunks);
+}
+
+function getResultTokensPerSecond(result: TestResult): number | null {
+  return (
+    result.tokensPerSecond ??
+    computeTokensPerSecond({
+      outputTokens: result.outputTokens,
+      firstChunkMs: getResultFirstChunkMs(result),
+      totalStreamMs: getResultTotalStreamMs(result),
+    })
+  );
+}
+
 async function logModelResults(
   displayName: string,
   providerName: string,
@@ -65,6 +85,9 @@ async function logModelResults(
         cacheCreationInputTokens: nonStreamingResult.cacheCreationInputTokens ?? null,
         cacheReadInputTokens: nonStreamingResult.cacheReadInputTokens ?? null,
         elapsedMs: nonStreamingResult.latencyMs ?? 0,
+        firstChunkMs: null,
+        totalStreamMs: null,
+        tokensPerSecond: null,
         streaming: false,
         userAgent: "provider-test",
         origin: null,
@@ -91,7 +114,10 @@ async function logModelResults(
         outputTokens: streamingResult.outputTokens ?? null,
         cacheCreationInputTokens: streamingResult.cacheCreationInputTokens ?? null,
         cacheReadInputTokens: streamingResult.cacheReadInputTokens ?? null,
-        elapsedMs: streamingResult.latencyMs ?? 0,
+        elapsedMs: getResultTotalStreamMs(streamingResult) ?? 0,
+        firstChunkMs: getResultFirstChunkMs(streamingResult),
+        totalStreamMs: getResultTotalStreamMs(streamingResult),
+        tokensPerSecond: getResultTokensPerSecond(streamingResult),
         streaming: true,
         streamingChunks: captureFullDetails ? streamingResult.streamingChunks : undefined,
         userAgent: "provider-test",
@@ -127,6 +153,9 @@ async function logModelResults(
         cacheCreationInputTokens: null,
         cacheReadInputTokens: null,
         elapsedMs: nonStreamingResult.latencyMs ?? 0,
+        firstChunkMs: null,
+        totalStreamMs: null,
+        tokensPerSecond: null,
         streaming: false,
         userAgent: "provider-test",
         origin: null,
@@ -153,7 +182,10 @@ async function logModelResults(
         outputTokens: streamingResult.outputTokens ?? null,
         cacheCreationInputTokens: null,
         cacheReadInputTokens: null,
-        elapsedMs: streamingResult.latencyMs ?? 0,
+        elapsedMs: getResultTotalStreamMs(streamingResult) ?? 0,
+        firstChunkMs: getResultFirstChunkMs(streamingResult),
+        totalStreamMs: getResultTotalStreamMs(streamingResult),
+        tokensPerSecond: getResultTokensPerSecond(streamingResult),
         streaming: true,
         streamingChunks: captureFullDetails ? streamingResult.streamingChunks : undefined,
         userAgent: "provider-test",