obol-ai 0.3.15 → 0.3.17

package/CHANGELOG.md

@@ -1,3 +1,9 @@
+## 0.3.16
+- Merge pull request #7 from jestersimpps/fix/stt-whisper-transcribe
+- fix: STT pipeline - add missing whisper_transcribe.py and fix media handler
+- Merge pull request #6 from jestersimpps/fix/status-instant-tool-labels
+- fix: replace async haiku status labels with instant sync formatToolCall
+
 ## 0.3.15
 - replace impulse with news system, exact datetime for analysis follow-ups
 - rewrite impulse prompt to sound like a friend, not an assistant

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "obol-ai",
-  "version": "0.3.15",
+  "version": "0.3.17",
   "description": "Self-evolving AI assistant that learns, remembers, and acts on its own. Persistent vector memory, self-rewriting personality, proactive heartbeats.",
   "main": "src/index.js",
   "bin": {

package/src/media/whisper_transcribe.py

@@ -0,0 +1,42 @@
+#!/usr/bin/env python3
+"""
+whisper_transcribe.py — Transcribe audio files using faster-whisper.
+
+Called by stt.js as: python3 whisper_transcribe.py <audio_file_path>
+
+Uses the 'tiny' model for fast CPU inference (~140MB).
+Outputs the transcribed text to stdout.
+Exits with code 1 on failure.
+"""
+
+import sys
+
+def transcribe(file_path):
+    """Transcribe an audio file and return the text."""
+    try:
+        from faster_whisper import WhisperModel
+    except ImportError:
+        print("ERROR: faster-whisper not installed. Run: pip3 install faster-whisper", file=sys.stderr)
+        sys.exit(1)
+
+    try:
+        model = WhisperModel("tiny", device="cpu", compute_type="int8")
+        segments, info = model.transcribe(file_path)
+        text = " ".join(segment.text.strip() for segment in segments).strip()
+        return text
+    except Exception as e:
+        print(f"ERROR: Transcription failed: {e}", file=sys.stderr)
+        sys.exit(1)
+
+
+if __name__ == "__main__":
+    if len(sys.argv) < 2:
+        print("Usage: python3 whisper_transcribe.py <audio_file_path>", file=sys.stderr)
+        sys.exit(1)
+
+    file_path = sys.argv[1]
+    result = transcribe(file_path)
+    if result:
+        print(result)
+    else:
+        sys.exit(1)

package/src/status.js

@@ -1,5 +1,4 @@
 const TERM_WIDTH = 25;
-const _toolDescriptionCache = new Map();
 
 function buildStatusHtml({ route, elapsed, toolStatus, title = 'OBOL' }) {
   const pad = Math.max(0, TERM_WIDTH - title.length - 3);
@@ -28,25 +27,10 @@ function buildStatusHtml({ route, elapsed, toolStatus, title = 'OBOL' }) {
   return `<pre>${lines.join('\n')}</pre>`;
 }
 
-function describeToolCall(client, toolName, inputSummary) {
-  const key = `${toolName}:${inputSummary}`;
-  const cached = _toolDescriptionCache.get(key);
-  if (cached) return Promise.resolve(cached);
-
-  return client.messages.create({
-    model: 'claude-haiku-4-5',
-    max_tokens: 30,
-    system: 'Describe this tool call in 3-8 words from the user\'s perspective. Present participle. No quotes, period, or emoji.',
-    messages: [{ role: 'user', content: `${toolName}: ${inputSummary}` }],
-  }).then(r => {
-    const desc = r.content[0]?.text?.trim() || null;
-    if (desc) _toolDescriptionCache.set(key, desc);
-    if (_toolDescriptionCache.size > 200) {
-      const first = _toolDescriptionCache.keys().next().value;
-      _toolDescriptionCache.delete(first);
-    }
-    return desc;
-  }).catch(() => null);
+function formatToolCall(toolName, inputSummary) {
+  if (!inputSummary) return toolName;
+  const truncated = inputSummary.length > 40 ? inputSummary.slice(0, 37) + '...' : inputSummary;
+  return `${toolName} "${truncated}"`;
 }
 
-module.exports = { buildStatusHtml, describeToolCall, TERM_WIDTH };
+module.exports = { buildStatusHtml, formatToolCall, TERM_WIDTH };

package/src/telegram/handlers/media.js

@@ -1,6 +1,6 @@
 const path = require('path');
 const { getTenant } = require('../../tenant');
-const { buildStatusHtml, describeToolCall } = require('../../status');
+const { buildStatusHtml, formatToolCall } = require('../../status');
 const media = require('../../media');
 const { sendHtml, startTyping, splitMessage } = require('../utils');
 const { MAX_MEDIA_SIZE, MEDIA_GROUP_DELAY_MS } = require('../constants');
@@ -43,9 +43,9 @@ async function processMediaItems(ctx, items, { config, allowedUsers, bot, create
         imageBlocks.push(media.bufferToImageBlock(item.buffer, item.fileInfo.mimeType));
       } else if (
         (item.fileInfo.mediaType === 'voice' || item.fileInfo.mediaType === 'audio') &&
-        tenant.toolPrefs?.get?.('speech_to_text')?.enabled === true
+        tenant.toolPrefs?.get?.('speech_to_text')?.enabled !== false
       ) {
-        const { transcribe } = require('../../stt');
+        const { transcribe } = require('../../media/stt');
         const transcription = await transcribe(savedPath);
         nonImageParts.push(transcription
           ? `[Voice message transcription: ${transcription}]`
@@ -80,11 +80,9 @@ async function processMediaItems(ctx, items, { config, allowedUsers, bot, create
       if (update.model) ri.model = update.model;
     };
     mediaChatCtx._onToolStart = (toolName, inputSummary) => {
-      status.setStatusText('Processing');
-      describeToolCall(tenant.claude.client, toolName, inputSummary).then(desc => {
-        if (desc) status.setStatusText(desc);
-      });
+      status.setStatusText(formatToolCall(toolName, inputSummary));
       status.start();
+      status.pushUpdate();
     };
     mediaChatCtx._onLockTimeout = () => {
       status.clear();

package/src/telegram/handlers/special.js

@@ -1,5 +1,5 @@
 const { getTenant } = require('../../tenant');
-const { describeToolCall } = require('../../status');
+const { formatToolCall } = require('../../status');
 const { sendHtml, startTyping, splitMessage } = require('../utils');
 const { createChatContext, createStatusTracker } = require('./text');
 
@@ -92,11 +92,9 @@ async function processSpecial(ctx, prompt, deps) {
       if (update.model) ri.model = update.model;
     };
     chatCtx._onToolStart = (toolName, inputSummary) => {
-      status.setStatusText('Processing');
-      describeToolCall(tenant.claude.client, toolName, inputSummary).then(desc => {
-        if (desc) status.setStatusText(desc);
-      });
+      status.setStatusText(formatToolCall(toolName, inputSummary));
       status.start();
+      status.pushUpdate();
     };
 
     const { text: response, usage, model } = await tenant.claude.chat(prompt, chatCtx);

package/src/telegram/handlers/text.js

@@ -1,6 +1,6 @@
 const { InlineKeyboard } = require('grammy');
 const { getTenant } = require('../../tenant');
-const { buildStatusHtml, describeToolCall } = require('../../status');
+const { buildStatusHtml, formatToolCall } = require('../../status');
 const { sendHtml, startTyping, splitMessage } = require('../utils');
 const { TEXT_BUFFER_GAP_MS, TEXT_BUFFER_MAX_PARTS, TEXT_BUFFER_MAX_CHARS, TEXT_BUFFER_THRESHOLD } = require('../constants');
 
@@ -133,6 +133,12 @@ function createStatusTracker(ctx, botName) {
       const html = buildStatusHtml({ route: routeInfo, elapsed, toolStatus: 'Formatting output', title });
       ctx.api.editMessageText(ctx.chat.id, statusMsgId, html, { parse_mode: 'HTML' }).catch(() => {});
     },
+    pushUpdate() {
+      if (!statusMsgId) return;
+      const elapsed = statusStart ? Math.round((Date.now() - statusStart) / 1000) : 0;
+      const html = buildStatusHtml({ route: routeInfo, elapsed, toolStatus: statusText, title });
+      ctx.api.editMessageText(ctx.chat.id, statusMsgId, html, { parse_mode: 'HTML', reply_markup: stopBtn }).catch(() => {});
+    },
     deleteMsg() {
       if (statusMsgId) ctx.api.deleteMessage(ctx.chat.id, statusMsgId).catch(() => {});
     },
@@ -185,11 +191,9 @@ async function processTextMessage(ctx, fullMessage, { config, allowedUsers, bot,
       if (update.model) ri.model = update.model;
     };
     chatContext._onToolStart = (toolName, inputSummary) => {
-      status.setStatusText('Processing');
-      describeToolCall(tenant.claude.client, toolName, inputSummary).then(desc => {
-        if (desc) status.setStatusText(desc);
-      });
+      status.setStatusText(formatToolCall(toolName, inputSummary));
       status.start();
+      status.pushUpdate();
     };
     chatContext._onLockTimeout = () => {
       status.clear();