getaimeter 0.1.6 → 0.1.8

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "getaimeter",
-  "version": "0.1.6",
+  "version": "0.1.8",
   "description": "Track your Claude AI usage across CLI, VS Code, and Desktop App. One command to start.",
   "bin": {
     "aimeter": "cli.js"

package/watcher.js

@@ -73,6 +73,7 @@ function extractNewUsage(filePath) {
 
   const usageEvents = [];
   let lineOffset = lastOffset;
+  let pendingThinkingChars = 0; // Track thinking chars from streaming progress messages
 
   for (const line of lines) {
     const trimmed = line.trim();
@@ -84,9 +85,28 @@ function extractNewUsage(filePath) {
 
     if (obj.type !== 'assistant' || !obj.message || !obj.message.usage) continue;
 
+    // Check for thinking content blocks (appear in streaming progress messages)
+    const contentBlocks = obj.message.content || [];
+    for (const block of contentBlocks) {
+      if (block.type === 'thinking' && block.thinking) {
+        pendingThinkingChars = Math.max(pendingThinkingChars, block.thinking.length);
+      }
+    }
+
+    // Skip streaming progress messages (stop_reason: null = not yet complete)
+    // Only count final responses with a real stop_reason
+    if (obj.message.stop_reason === null || obj.message.stop_reason === undefined) continue;
+
     const u = obj.message.usage;
     const model = obj.message.model || 'unknown';
 
+    // Estimate thinking tokens: ~4 chars per token (conservative estimate)
+    // The API doesn't separate thinking_tokens in the JSONL usage field
+    const estimatedThinkingTokens = pendingThinkingChars > 0
+      ? Math.ceil(pendingThinkingChars / 4)
+      : 0;
+    pendingThinkingChars = 0; // Reset for next turn
+
     // Build dedup hash — include line offset for uniqueness
     const hash = crypto.createHash('md5')
       .update(`${filePath}:${lineOffset}:${model}:${u.input_tokens || 0}:${u.output_tokens || 0}`)
@@ -100,7 +120,7 @@ function extractNewUsage(filePath) {
       source:           detectSource(filePath),
       inputTokens:      u.input_tokens               || 0,
       outputTokens:     u.output_tokens              || 0,
-      thinkingTokens:   u.thinking_tokens            || 0,
+      thinkingTokens:   estimatedThinkingTokens,
       cacheReadTokens:  u.cache_read_input_tokens    || 0,
       cacheWriteTokens: u.cache_creation_input_tokens || 0,
     });