docs: add cost estimation and tool timing features to AI SDK

Co-Authored-By: open-agents-bot[bot] <260704009+open-agents-bot[bot]@users.noreply.github.com>

Author: HugoRCD

.vercel/project.json

@@ -0,0 +1,5 @@
+{
+  "orgId": "team_X7ikPPn4YpZ4APwufE36i53l",
+  "projectId": "prj_GO3fYDyQKzkoreOaBBgTIBRZVvt0",
+  "projectName": "evlog-docs"
+}

apps/docs/app/components/features/FeatureAiSdk.vue

@@ -17,6 +17,8 @@ const pills = [
   { label: 'Token tracking', icon: 'i-lucide-coins' },
   { label: 'Tool calls', icon: 'i-lucide-wrench' },
   { label: 'Streaming metrics', icon: 'i-lucide-gauge' },
+  { label: 'Cost estimation', icon: 'i-lucide-dollar-sign' },
+  { label: 'Tool timing', icon: 'i-lucide-timer' },
 ]
 
 const benefits = [
@@ -28,12 +30,17 @@ const benefits = [
   {
     icon: 'i-lucide-box',
     title: 'Works with everything',
-    text: 'generateText, streamText, ToolLoopAgent, generateObject.',
+    text: 'generateText, streamText, ToolLoopAgent, embed, multi-step agents.',
   },
   {
     icon: 'i-lucide-bar-chart-3',
     title: 'Cost and performance',
-    text: 'Token usage, cache hits, time to first chunk, tokens per second.',
+    text: 'Token usage, cache hits, cost estimation, time to first chunk, tokens per second.',
+  },
+  {
+    icon: 'i-lucide-timer',
+    title: 'Telemetry integration',
+    text: 'Per-tool execution timing, success/failure tracking, and total generation wall time.',
   },
 ]
 
@@ -175,11 +182,17 @@ function setView(view: 'without' | 'with') {
 
           <div class="px-5 pt-4 pb-3 font-mono text-xs sm:text-sm leading-relaxed overflow-x-auto border-b border-muted/50">
             <!-- eslint-disable vue/multiline-html-element-content-newline -->
-            <pre><code><span class="text-violet-400">const</span> ai = <span class="text-amber-400">createAILogger</span>(log)
+            <pre><code><span class="text-violet-400">const</span> ai = <span class="text-amber-400">createAILogger</span>(log, {
+  <span class="text-sky-400">cost</span>: { <span class="text-emerald-400">'claude-sonnet-4.6'</span>: { <span class="text-sky-400">input</span>: <span class="text-pink-400">3</span>, <span class="text-sky-400">output</span>: <span class="text-pink-400">15</span> } },
+})
 
 <span class="text-violet-400">const</span> result = <span class="text-amber-400">streamText</span>({
   <span class="text-sky-400">model</span>: ai.<span class="text-amber-400">wrap</span>(<span class="text-emerald-400">'anthropic/claude-sonnet-4.6'</span>),
   messages,
+  <span class="text-sky-400">experimental_telemetry</span>: {
+    <span class="text-sky-400">isEnabled</span>: <span class="text-violet-400">true</span>,
+    <span class="text-sky-400">integrations</span>: [<span class="text-amber-400">createEvlogIntegration</span>(ai)],
+  },
 })</code></pre>
             <!-- eslint-enable -->
           </div>
@@ -248,8 +261,12 @@ function setView(view: 'without' | 'with') {
                   <span class="text-amber-400"> ["searchWeb", "queryDB"]</span>
                 </div>
                 <div>
-                  <span class="text-sky-400">ai.steps</span><span class="text-dimmed">:</span>
-                  <span class="text-pink-400"> 3</span>
+                  <span class="text-sky-400">ai.tools</span><span class="text-dimmed">:</span>
+                  <span class="text-amber-400"> [{name: "searchWeb", durationMs: 150, ...}]</span>
+                </div>
+                <div>
+                  <span class="text-sky-400">ai.estimatedCost</span><span class="text-dimmed">:</span>
+                  <span class="text-emerald-400"> 0.022</span>
                 </div>
                 <div>
                   <span class="text-sky-400">ai.msToFirstChunk</span><span class="text-dimmed">:</span>

apps/docs/content/0.landing.md

@@ -101,7 +101,7 @@ Wide events and structured errors for TypeScript. One log per request, full cont
   Make AI calls :br observable
 
   #description
-  Your AI endpoints are black boxes. You don't know how many tokens each request burns, which tools the model called, or how fast the stream was. Wrap your model with one line and every call is captured into the wide event. Cost tracking, tool visibility, streaming performance, cache hits, reasoning tokens.
+  Your AI endpoints are black boxes. You don't know how many tokens each request burns, which tools the model called, or how fast the stream was. Wrap your model with one line and every call is captured into the wide event. Cost estimation, tool execution timing, streaming performance, cache hits, reasoning tokens, and multi-step agent breakdowns.
   :::
 
   :::features-feature-performance