@tonyclaw/agent-inspector 2.0.28 → 2.0.29

package/.output/server/index.mjs

@@ -35,103 +35,103 @@ const headers = ((m) => function headersRouteRule(event) {
   }
 });
 const assets = {
-  "/assets/index-B9_VaAWl.js": {
-    "type": "text/javascript; charset=utf-8",
-    "etag": '"74-wegwMejcMaCe2qDEma7y6qUh/6U"',
-    "mtime": "2026-06-30T03:21:19.568Z",
-    "size": 116,
-    "path": "../public/assets/index-B9_VaAWl.js"
-  },
   "/assets/alibaba-TTwafVwX.svg": {
     "type": "image/svg+xml",
     "etag": '"171b-6dyV5K8QjiaY35sN9qNprh9zDIs"',
-    "mtime": "2026-06-30T03:21:19.568Z",
+    "mtime": "2026-06-30T09:01:48.556Z",
     "size": 5915,
     "path": "../public/assets/alibaba-TTwafVwX.svg"
   },
-  "/assets/CompareDrawer-CicCP3Hb.js": {
+  "/assets/index-BhQGSdhG.js": {
     "type": "text/javascript; charset=utf-8",
-    "etag": '"49fe-CmOMF3QSwAcX6NWjQdoZrIdyoIw"',
-    "mtime": "2026-06-30T03:21:19.570Z",
-    "size": 18942,
-    "path": "../public/assets/CompareDrawer-CicCP3Hb.js"
+    "etag": '"74-1e7gU3IOVaEjblxPQfgH61GRjlw"',
+    "mtime": "2026-06-30T09:01:48.556Z",
+    "size": 116,
+    "path": "../public/assets/index-BhQGSdhG.js"
   },
-  "/assets/index-DsiKfWCp.css": {
-    "type": "text/css; charset=utf-8",
-    "etag": '"1a3d5-pnLJEAiyUUKlRw5OUnvwf1dCQeI"',
-    "mtime": "2026-06-30T03:21:19.568Z",
-    "size": 107477,
-    "path": "../public/assets/index-DsiKfWCp.css"
+  "/assets/CompareDrawer-DcqxlgSQ.js": {
+    "type": "text/javascript; charset=utf-8",
+    "etag": '"49fe-+D2I+/9sd7EYksKARn3HVY2eJf0"',
+    "mtime": "2026-06-30T09:01:48.556Z",
+    "size": 18942,
+    "path": "../public/assets/CompareDrawer-DcqxlgSQ.js"
   },
   "/assets/minimax-BPMzvuL-.jpeg": {
     "type": "image/jpeg",
     "etag": '"1b06-IwivU89ko5UTMUM1/t7hn4sQK9A"',
-    "mtime": "2026-06-30T03:21:19.568Z",
+    "mtime": "2026-06-30T09:01:48.556Z",
     "size": 6918,
     "path": "../public/assets/minimax-BPMzvuL-.jpeg"
   },
-  "/assets/qwen-CONDcHqt.png": {
-    "type": "image/png",
-    "etag": '"572c3-cdJAPaHdOvFCGzuaQjagdgOu6XE"',
-    "mtime": "2026-06-30T03:21:19.568Z",
-    "size": 357059,
-    "path": "../public/assets/qwen-CONDcHqt.png"
+  "/assets/index-CwlHPmgL.css": {
+    "type": "text/css; charset=utf-8",
+    "etag": '"1a3f0-E2f0kOdSBi1RhjoK38+CSij/apg"',
+    "mtime": "2026-06-30T09:01:48.556Z",
+    "size": 107504,
+    "path": "../public/assets/index-CwlHPmgL.css"
   },
-  "/assets/ReplayDialog-Bc8q3ujm.js": {
+  "/assets/ReplayDialog-CBcnPgx1.js": {
     "type": "text/javascript; charset=utf-8",
-    "etag": '"318c-QS16LlsGhExlRZzuuDmYzqWLBN0"',
-    "mtime": "2026-06-30T03:21:19.570Z",
+    "etag": '"318c-Wq35vSCO67O3pnw6riReoXT9gOg"',
+    "mtime": "2026-06-30T09:01:48.557Z",
     "size": 12684,
-    "path": "../public/assets/ReplayDialog-Bc8q3ujm.js"
+    "path": "../public/assets/ReplayDialog-CBcnPgx1.js"
   },
-  "/assets/zhipuai-BPNAnxo-.svg": {
-    "type": "image/svg+xml",
-    "etag": '"2bf8-hNaLCTi89nOFCsIIfWpP/jrfo0s"',
-    "mtime": "2026-06-30T03:21:19.568Z",
-    "size": 11256,
-    "path": "../public/assets/zhipuai-BPNAnxo-.svg"
+  "/assets/RequestAnatomy-DSgSqCYt.js": {
+    "type": "text/javascript; charset=utf-8",
+    "etag": '"4485-rBai6wnmYHf662t3omQFALcwHSw"',
+    "mtime": "2026-06-30T09:01:48.557Z",
+    "size": 17541,
+    "path": "../public/assets/RequestAnatomy-DSgSqCYt.js"
   },
-  "/assets/main-Brj0Gn91.js": {
+  "/assets/ResponseView-CjqzBSoF.js": {
     "type": "text/javascript; charset=utf-8",
-    "etag": '"51726-mwq2VuKeQG//deWEjQSFZIL9ZWo"',
-    "mtime": "2026-06-30T03:21:19.568Z",
-    "size": 333606,
-    "path": "../public/assets/main-Brj0Gn91.js"
+    "etag": '"30d0-13nPwe9jIZSSl43vTEhXlswXwn8"',
+    "mtime": "2026-06-30T09:01:48.557Z",
+    "size": 12496,
+    "path": "../public/assets/ResponseView-CjqzBSoF.js"
   },
-  "/assets/RequestAnatomy-Coyy3zcH.js": {
+  "/assets/StreamingChunkSequence-DdBBKolI.js": {
     "type": "text/javascript; charset=utf-8",
-    "etag": '"4485-KFNCPYVKcx+a4jW1uo7q7Az4T5c"',
-    "mtime": "2026-06-30T03:21:19.570Z",
-    "size": 17541,
-    "path": "../public/assets/RequestAnatomy-Coyy3zcH.js"
+    "etag": '"d57-9H1QMwesXGWHkYNTPJ/9zDoslX0"',
+    "mtime": "2026-06-30T09:01:48.557Z",
+    "size": 3415,
+    "path": "../public/assets/StreamingChunkSequence-DdBBKolI.js"
   },
-  "/assets/_sessionId-9rEF0uSE.js": {
+  "/assets/_sessionId-wctOoGh_.js": {
     "type": "text/javascript; charset=utf-8",
-    "etag": '"d2-+KgPaqhF/ImO1Zb9tZi0sMB8OPI"',
-    "mtime": "2026-06-30T03:21:19.568Z",
+    "etag": '"d2-fiIvvYyX9jF4fd1W7HpowwQ2Zj4"',
+    "mtime": "2026-06-30T09:01:48.556Z",
     "size": 210,
-    "path": "../public/assets/_sessionId-9rEF0uSE.js"
+    "path": "../public/assets/_sessionId-wctOoGh_.js"
   },
-  "/assets/ResponseView-BWJzgh0c.js": {
+  "/assets/main-B6OLZCp9.js": {
     "type": "text/javascript; charset=utf-8",
-    "etag": '"30d0-XQuX3LcRTXyoZYFeRDp9J2gMpd0"',
-    "mtime": "2026-06-30T03:21:19.570Z",
-    "size": 12496,
-    "path": "../public/assets/ResponseView-BWJzgh0c.js"
+    "etag": '"51726-MaBaigWv6OLiiYg6fusFyN758Hw"',
+    "mtime": "2026-06-30T09:01:48.557Z",
+    "size": 333606,
+    "path": "../public/assets/main-B6OLZCp9.js"
   },
-  "/assets/StreamingChunkSequence-BSC3uuE5.js": {
-    "type": "text/javascript; charset=utf-8",
-    "etag": '"d57-1GnBSzcha+n1obLlcv4lwyBhPf4"',
-    "mtime": "2026-06-30T03:21:19.570Z",
-    "size": 3415,
-    "path": "../public/assets/StreamingChunkSequence-BSC3uuE5.js"
+  "/assets/zhipuai-BPNAnxo-.svg": {
+    "type": "image/svg+xml",
+    "etag": '"2bf8-hNaLCTi89nOFCsIIfWpP/jrfo0s"',
+    "mtime": "2026-06-30T09:01:48.556Z",
+    "size": 11256,
+    "path": "../public/assets/zhipuai-BPNAnxo-.svg"
+  },
+  "/assets/qwen-CONDcHqt.png": {
+    "type": "image/png",
+    "etag": '"572c3-cdJAPaHdOvFCGzuaQjagdgOu6XE"',
+    "mtime": "2026-06-30T09:01:48.556Z",
+    "size": 357059,
+    "path": "../public/assets/qwen-CONDcHqt.png"
   },
-  "/assets/ProxyViewerContainer-BccuA6p5.js": {
+  "/assets/ProxyViewerContainer-D650kQlv.js": {
     "type": "text/javascript; charset=utf-8",
-    "etag": '"af111-eqyso0QJHFS7+IdUoP/1PcawtQE"',
-    "mtime": "2026-06-30T03:21:19.570Z",
-    "size": 717073,
-    "path": "../public/assets/ProxyViewerContainer-BccuA6p5.js"
+    "etag": '"af8bd-kLHfI4KjZ80Xpnq5/VoFdzyBH10"',
+    "mtime": "2026-06-30T09:01:48.557Z",
+    "size": 719037,
+    "path": "../public/assets/ProxyViewerContainer-D650kQlv.js"
   }
 };
 function readAsset(id) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tonyclaw/agent-inspector",
-  "version": "2.0.28",
+  "version": "2.0.29",
   "type": "module",
   "description": "Agent observability and knowledge capture layer for AI coding tools.",
   "license": "MIT",

package/src/components/ProxyViewer.tsx

@@ -783,6 +783,9 @@ export function ProxyViewer({
               <Plus className="size-4 shrink-0 text-muted-foreground/70" aria-hidden="true" />
               <McpLogo className="size-10 shrink-0" />
             </span>
+            <span className="max-w-[calc(100vw-7rem)] text-sm font-semibold text-muted-foreground">
+              Local First. Evidence First.
+            </span>
             <McpReadyBadge />
           </h1>
           <div className="justify-self-end">

package/src/components/providers/ProviderCard.tsx

@@ -87,6 +87,16 @@ function hasHeaderDetails(headers: Record<string, string> | undefined): boolean
   return formatHeaders(headers) !== undefined;
 }
 
+function formatElapsed(ms: number): string {
+  if (ms < 1000) return `${ms}ms`;
+  return `${(ms / 1000).toFixed(1)}s`;
+}
+
+function formatTokenRate(tokensPerSecond: number): string {
+  if (tokensPerSecond >= 10) return `${tokensPerSecond.toFixed(0)}/s`;
+  return `${tokensPerSecond.toFixed(1)}/s`;
+}
+
 function hasDiagnosticDetails(result: TestResult): boolean {
   const debug = result.debug;
   return (
@@ -255,6 +265,7 @@ function TestStatus({ result }: { result: ProviderTestState }): JSX.Element {
   }
   if (result.success) {
     const tokenParts: ReactNode[] = [];
+    const timingParts: ReactNode[] = [];
     if (result.inputTokens !== undefined) {
       tokenParts.push(<span key="in">{result.inputTokens} in</span>);
     }
@@ -275,11 +286,29 @@ function TestStatus({ result }: { result: ProviderTestState }): JSX.Element {
         </span>,
       );
     }
+    if (result.firstChunkMs !== undefined && result.firstChunkMs !== null) {
+      timingParts.push(<span key="first">first {formatElapsed(result.firstChunkMs)}</span>);
+    }
+    if (result.totalStreamMs !== undefined && result.totalStreamMs !== null) {
+      timingParts.push(<span key="total">total {formatElapsed(result.totalStreamMs)}</span>);
+    }
+    if (
+      result.tokensPerSecond !== undefined &&
+      result.tokensPerSecond !== null &&
+      result.tokensPerSecond > 0
+    ) {
+      timingParts.push(<span key="rate">{formatTokenRate(result.tokensPerSecond)} tok/s</span>);
+    }
     const displayTokens: ReactNode[] = [];
     for (let i = 0; i < tokenParts.length; i++) {
       if (i > 0) displayTokens.push(", ");
       displayTokens.push(tokenParts[i]);
     }
+    const displayTiming: ReactNode[] = [];
+    for (let i = 0; i < timingParts.length; i++) {
+      if (i > 0) displayTiming.push(", ");
+      displayTiming.push(timingParts[i]);
+    }
     return (
       <TooltipProvider>
         <Tooltip>
@@ -290,6 +319,9 @@ function TestStatus({ result }: { result: ProviderTestState }): JSX.Element {
               {tokenParts.length > 0 && (
                 <span className="text-muted-foreground">({displayTokens})</span>
               )}
+              {timingParts.length > 0 && (
+                <span className="hidden text-muted-foreground lg:inline">({displayTiming})</span>
+              )}
             </div>
           </TooltipTrigger>
           <TooltipContent>Connection test passed</TooltipContent>

package/src/components/proxy-viewer/LogEntryHeader.tsx

@@ -11,6 +11,7 @@ import {
   Copy,
   FileDiff,
   FileTerminal,
+  Gauge,
   Globe,
   History,
   Loader2,
@@ -45,6 +46,11 @@ function formatElapsed(ms: number): string {
   return `${(ms / 1000).toFixed(1)}s`;
 }
 
+function formatTokenRate(tokensPerSecond: number): string {
+  if (tokensPerSecond >= 10) return `${tokensPerSecond.toFixed(0)}/s`;
+  return `${tokensPerSecond.toFixed(1)}/s`;
+}
+
 /**
  * Inline trend indicator: small arrow (green up / red down) plus the absolute
  * delta in compact form. Returns `null` when there is no trend to display.
@@ -148,6 +154,11 @@ export const LogEntryHeader = memo(function ({
     log.elapsedMs > slowResponseThresholdSeconds * 1000;
 
   const hasTokens = log.inputTokens !== null || log.outputTokens !== null;
+  const firstChunkMs = log.firstChunkMs ?? null;
+  const tokensPerSecond = log.tokensPerSecond ?? null;
+  const firstChunkLabel = firstChunkMs === null ? null : formatElapsed(firstChunkMs);
+  const tokenRateLabel =
+    tokensPerSecond === null || tokensPerSecond <= 0 ? null : formatTokenRate(tokensPerSecond);
 
   return (
     <TooltipProvider>
@@ -238,11 +249,37 @@ export const LogEntryHeader = memo(function ({
                 ? `Slow response: ${formatElapsed(log.elapsedMs)} exceeds ${formatElapsed(
                     slowResponseThresholdSeconds * 1000,
                   )}`
-                : "Elapsed response time"}
+                : log.streaming
+                  ? "Total streaming response time"
+                  : "Elapsed response time"}
             </TooltipContent>
           </Tooltip>
         )}
 
+        {/* Streaming timing */}
+        {log.streaming && firstChunkLabel !== null && (
+          <Tooltip>
+            <TooltipTrigger asChild>
+              <span className="flex items-center gap-1 text-xs text-cyan-400 shrink-0">
+                <Radio className="size-3" />
+                <span className="font-mono tabular-nums">first {firstChunkLabel}</span>
+              </span>
+            </TooltipTrigger>
+            <TooltipContent>Time to first streaming chunk</TooltipContent>
+          </Tooltip>
+        )}
+        {log.streaming && tokenRateLabel !== null && (
+          <Tooltip>
+            <TooltipTrigger asChild>
+              <span className="flex items-center gap-1 text-xs text-emerald-400 shrink-0">
+                <Gauge className="size-3" />
+                <span className="font-mono tabular-nums">{tokenRateLabel}</span>
+              </span>
+            </TooltipTrigger>
+            <TooltipContent>Output token throughput after first chunk</TooltipContent>
+          </Tooltip>
+        )}
+
         {/* Token counts */}
         {hasTokens && (
           <span className="flex items-center gap-1 text-xs shrink-0">

package/src/lib/providerTestContract.ts

@@ -45,6 +45,9 @@ export const ProviderTestResultSchema = z.object({
   cacheCreationInputTokens: z.number().optional(),
   cacheReadInputTokens: z.number().optional(),
   latencyMs: z.number().optional(),
+  firstChunkMs: z.number().nullable().optional(),
+  totalStreamMs: z.number().nullable().optional(),
+  tokensPerSecond: z.number().nullable().optional(),
   content: z.array(ProviderTestContentSchema).optional(),
   rawResponse: z.string().optional(),
   streaming: z.boolean().optional(),

package/src/lib/sessionInfoContract.ts

@@ -21,6 +21,9 @@ export const SessionLogSummarySchema = z.object({
   hasError: z.boolean(),
   error: z.string().nullable(),
   latencyMs: z.number().nullable(),
+  firstChunkMs: z.number().nullable().optional(),
+  totalStreamMs: z.number().nullable().optional(),
+  tokensPerSecond: z.number().nullable().optional(),
   tokens: z.object({
     input: z.number().nullable(),
     output: z.number().nullable(),

package/src/mcp/toolHandlers.ts

@@ -132,6 +132,9 @@ export function buildLogSummary(log: CapturedLog) {
     isStreaming: log.streaming,
     hasError,
     latencyMs: log.elapsedMs,
+    firstChunkMs: log.firstChunkMs ?? null,
+    totalStreamMs: log.totalStreamMs ?? null,
+    tokensPerSecond: log.tokensPerSecond ?? null,
     tokens: {
       input: log.inputTokens,
       output: log.outputTokens,

package/src/proxy/evidenceAnalysis.ts

@@ -318,6 +318,11 @@ function logDetails(log: SessionLogSummary): string {
     `streaming=${log.isStreaming ? "yes" : "no"}`,
     `latencyMs=${formatNullable(log.latencyMs)}`,
   ];
+  if (log.isStreaming) {
+    parts.push(`firstChunkMs=${formatNullable(log.firstChunkMs ?? null)}`);
+    parts.push(`totalStreamMs=${formatNullable(log.totalStreamMs ?? null)}`);
+    parts.push(`tokensPerSecond=${formatNullable(log.tokensPerSecond ?? null)}`);
+  }
   if (log.error !== null) {
     parts.push(`error=${log.error}`);
   }

package/src/proxy/evidenceExporter.ts

@@ -112,14 +112,19 @@ function buildMarkdown(document: EvidenceDocument): string {
       lines.push("No compact logs were available.", "");
     } else {
       lines.push(
-        "| ID | Status | Model | Latency | Path | Error |",
-        "| --- | --- | --- | --- | --- | --- |",
+        "| ID | Status | Model | Latency | Stream | Path | Error |",
+        "| --- | --- | --- | --- | --- | --- | --- |",
       );
       for (const log of session.latestLogs) {
+        const streamTiming = log.isStreaming
+          ? `first=${log.firstChunkMs ?? "n/a"}ms total=${log.totalStreamMs ?? "n/a"}ms tps=${
+              log.tokensPerSecond ?? "n/a"
+            }`
+          : "n/a";
         lines.push(
           `| ${log.id} | ${log.status ?? "n/a"} | ${log.model ?? "n/a"} | ${
             log.latencyMs ?? "n/a"
-          } | ${log.path} | ${log.hasError ? "yes" : "no"} |`,
+          } | ${streamTiming} | ${log.path} | ${log.hasError ? "yes" : "no"} |`,
         );
       }
       lines.push("");

package/src/proxy/handler.ts

@@ -142,12 +142,14 @@ function handleStreamingResponse(
   const capture = new RawStreamCapture();
   const decoder = new TextDecoder();
   let finalizationScheduled = false;
+  let firstChunkMs: number | null = null;
 
   const scheduleStreamingFinalization = (type: "streaming" | "stream-abort"): void => {
     if (finalizationScheduled) return;
     finalizationScheduled = true;
 
-    const elapsedMs = Date.now() - startTime;
+    const totalStreamMs = Date.now() - startTime;
+    const elapsedMs = totalStreamMs;
     const rawStream = capture.snapshot();
     switch (type) {
       case "streaming":
@@ -159,6 +161,8 @@ function handleStreamingResponse(
           responseStatus: upstreamRes.status,
           rawStream,
           collectStreamingChunks,
+          firstChunkMs,
+          totalStreamMs,
         });
         return;
       case "stream-abort":
@@ -170,6 +174,8 @@ function handleStreamingResponse(
           rawStream,
           hasChunks: capture.hasChunks,
           collectStreamingChunks,
+          firstChunkMs,
+          totalStreamMs,
         });
         return;
     }
@@ -177,6 +183,7 @@ function handleStreamingResponse(
 
   const transform = new TransformStream<Uint8Array, Uint8Array>({
     async transform(chunk, controller) {
+      if (firstChunkMs === null) firstChunkMs = Date.now() - startTime;
       controller.enqueue(chunk);
       await capture.append(decoder.decode(chunk, { stream: true }));
     },

package/src/proxy/logFinalizer.ts

@@ -8,6 +8,7 @@ import type { RawStreamSource } from "./rawStreamCapture";
 import type { CapturedLog } from "./schemas";
 import { getSessionProcess, isSessionProcessAvailable } from "./sessionProcess";
 import { finalizeLogUpdate } from "./store";
+import { computeTokensPerSecond } from "./streamTiming";
 
 type BaseFinalizeLogJob = {
   log: CapturedLog;
@@ -26,6 +27,8 @@ export type FinalizeStreamingLogJob = BaseFinalizeLogJob & {
   responseStatus: number;
   rawStream: string | RawStreamSource;
   collectStreamingChunks: boolean;
+  firstChunkMs: number | null;
+  totalStreamMs: number;
 };
 
 export type FinalizeStreamAbortLogJob = BaseFinalizeLogJob & {
@@ -33,6 +36,8 @@ export type FinalizeStreamAbortLogJob = BaseFinalizeLogJob & {
   rawStream: string | RawStreamSource;
   hasChunks: boolean;
   collectStreamingChunks: boolean;
+  firstChunkMs: number | null;
+  totalStreamMs: number;
 };
 
 export type FinalizeLogJob =
@@ -73,6 +78,9 @@ export function buildFileLogEntry(log: CapturedLog, upstreamUrl: string): Record
     cacheCreationInputTokens: log.cacheCreationInputTokens,
     cacheReadInputTokens: log.cacheReadInputTokens,
     elapsedMs: log.elapsedMs,
+    firstChunkMs: log.firstChunkMs,
+    totalStreamMs: log.totalStreamMs,
+    tokensPerSecond: log.tokensPerSecond,
     streaming: log.streaming,
     userAgent: log.userAgent,
     origin: log.origin,
@@ -119,10 +127,31 @@ function finalizeWithError(
   log.responseStatus = log.responseStatus ?? fallbackStatus;
   log.responseText = log.responseText ?? fallbackResponseText;
   log.elapsedMs = job.elapsedMs;
+  applyOptionalStreamingTiming(log, job);
   log.error = message;
   return { log, upstreamUrl: job.upstreamUrl, error: message, cleanupRawStreamPath };
 }
 
+function applyOptionalStreamingTiming(log: CapturedLog, job: FinalizeLogJob): void {
+  switch (job.type) {
+    case "non-streaming":
+      log.firstChunkMs = null;
+      log.totalStreamMs = null;
+      log.tokensPerSecond = null;
+      return;
+    case "streaming":
+    case "stream-abort":
+      log.firstChunkMs = job.firstChunkMs;
+      log.totalStreamMs = job.totalStreamMs;
+      log.tokensPerSecond = computeTokensPerSecond({
+        outputTokens: log.outputTokens,
+        firstChunkMs: job.firstChunkMs,
+        totalStreamMs: job.totalStreamMs,
+      });
+      return;
+  }
+}
+
 function finalizeNonStreaming(
   job: FinalizeNonStreamingLogJob,
   log: CapturedLog,
@@ -136,6 +165,9 @@ function finalizeNonStreaming(
     const tokens = formatHandler.extractTokens(job.responseBody);
 
     log.elapsedMs = job.elapsedMs;
+    log.firstChunkMs = null;
+    log.totalStreamMs = null;
+    log.tokensPerSecond = null;
     log.responseStatus = job.responseStatus;
     log.responseText = job.responseBody;
     log.inputTokens = tokens.inputTokens;
@@ -238,6 +270,7 @@ function finalizeStreaming(job: FinalizeStreamingLogJob, log: CapturedLog): Fina
   const oversized = getOversizedRawStream(job.rawStream);
   if (oversized !== null) {
     log.elapsedMs = job.elapsedMs;
+    applyOptionalStreamingTiming(log, job);
     log.responseStatus = job.responseStatus;
     log.responseText = oversizedRawStreamText(oversized);
     return {
@@ -262,6 +295,7 @@ function finalizeStreaming(job: FinalizeStreamingLogJob, log: CapturedLog): Fina
       log.model ?? undefined,
       job.collectStreamingChunks,
     );
+    applyOptionalStreamingTiming(log, job);
     persistStreamingChunks(log);
 
     return { log, upstreamUrl: job.upstreamUrl, error: null, cleanupRawStreamPath };
@@ -294,6 +328,7 @@ function finalizeStreamAbort(job: FinalizeStreamAbortLogJob, log: CapturedLog):
   let cleanupRawStreamPath: string | null = null;
   try {
     log.elapsedMs = job.elapsedMs;
+    applyOptionalStreamingTiming(log, job);
     if (job.hasChunks && formatHandler !== null) {
       const oversized = getOversizedRawStream(job.rawStream);
       if (oversized !== null) {
@@ -314,6 +349,7 @@ function finalizeStreamAbort(job: FinalizeStreamAbortLogJob, log: CapturedLog):
         log.model ?? undefined,
         job.collectStreamingChunks,
       );
+      applyOptionalStreamingTiming(log, job);
       persistStreamingChunks(log);
     } else {
       log.responseText = "Client aborted";

package/src/proxy/schemas.ts

@@ -40,6 +40,9 @@ export const CapturedLogSchema = z.object({
   cacheCreationInputTokens: z.number().nullable(),
   cacheReadInputTokens: z.number().nullable(),
   elapsedMs: z.number().nullable(),
+  firstChunkMs: z.number().nullable().optional(),
+  totalStreamMs: z.number().nullable().optional(),
+  tokensPerSecond: z.number().nullable().optional(),
   streaming: z.boolean(),
   userAgent: z.string().nullable(),
   origin: z.string().nullable(),

package/src/proxy/sessionInfo.ts

@@ -55,6 +55,9 @@ export function buildSessionLogSummary(log: CapturedLog): SessionLogSummary {
     hasError: hasLogError(log),
     error: truncateError(log.error),
     latencyMs: log.elapsedMs,
+    firstChunkMs: log.firstChunkMs ?? null,
+    totalStreamMs: log.totalStreamMs ?? null,
+    tokensPerSecond: log.tokensPerSecond ?? null,
     tokens: {
       input: log.inputTokens,
       output: log.outputTokens,

package/src/proxy/store.ts

@@ -233,6 +233,9 @@ export async function createLog(
     cacheCreationInputTokens: null,
     cacheReadInputTokens: null,
     elapsedMs: null,
+    firstChunkMs: null,
+    totalStreamMs: null,
+    tokensPerSecond: null,
     streaming: false,
     userAgent,
     origin,

package/src/proxy/streamTiming.ts

@@ -0,0 +1,31 @@
+import type { StreamingChunk } from "./schemas";
+
+export function computeTokensPerSecond({
+  outputTokens,
+  firstChunkMs,
+  totalStreamMs,
+}: {
+  outputTokens: number | null | undefined;
+  firstChunkMs: number | null | undefined;
+  totalStreamMs: number | null | undefined;
+}): number | null {
+  if (outputTokens === null || outputTokens === undefined || outputTokens <= 0) return null;
+  if (totalStreamMs === null || totalStreamMs === undefined || totalStreamMs <= 0) return null;
+
+  const generationMs =
+    firstChunkMs === null || firstChunkMs === undefined
+      ? totalStreamMs
+      : totalStreamMs - firstChunkMs;
+  if (!Number.isFinite(generationMs) || generationMs <= 0) return null;
+
+  return Number((outputTokens / (generationMs / 1000)).toFixed(2));
+}
+
+export function firstChunkMsFromChunks(
+  chunks: readonly StreamingChunk[] | undefined,
+): number | null {
+  const first = chunks?.[0];
+  if (first === undefined) return null;
+  if (!Number.isFinite(first.timestamp) || first.timestamp < 0) return null;
+  return Math.floor(first.timestamp);
+}