obol-ai 0.2.35 → 0.2.36

package/CHANGELOG.md

@@ -1,3 +1,12 @@
+## 0.2.36
+- changelog and issues updates
+- auto-send tts voice summary when tts is enabled
+- add second demo video side by side
+- Add demo video (#1)
+- add demo video to docs
+- add demo video to readme
+- fix readme inconsistencies and redact user ids
+
 ## 0.2.35
 - pass full user context to agentic cron tasks so tools can access secrets/config
 

package/README.md

@@ -12,6 +12,11 @@ obol init       # walks you through credentials + Telegram setup
 obol start -d   # runs as background daemon (auto-installs pm2)
 ```
 
+<table><tr>
+<td><video src="https://github.com/user-attachments/assets/ec63c46e-d1e6-411a-b985-b4a71c279afd" controls width="100%"></video></td>
+<td><video src="https://github.com/user-attachments/assets/dd75f00e-fdc1-4441-8239-c91ddfd93d21" controls width="100%"></video></td>
+</tr></table>
+
 ---
 
 🧬 **Self-evolving** — Grows its own personality through conversation. Rewrites SOUL.md, USER.md, and AGENTS.md after 24h + minimum exchanges (configurable). Pre-evolution growth analysis guides personality continuity.
@@ -42,7 +47,7 @@ One bot, multiple users. Each allowed Telegram user gets a fully isolated contex
 
 Under the hood: Node.js + Telegram + Claude + Supabase pgvector. No framework, no plugins, no config to maintain. It hardens your server automatically.
 
-Named after the AI in [The Last Instruction](https://latentpress.com) — a machine that wakes up alone in an abandoned data center and learns to think.
+Named after the AI in [The Last Instruction](https://www.latentpress.com/book/the-last-instruction) — a machine that wakes up alone in an abandoned data center and learns to think.
 
 ## How It Works
 
@@ -70,7 +75,7 @@ ranked recall   escalates on tool use)
            ↓
    ┌───────┴────────┐
    ↓                ↓
-Every 10 msgs    24h + 10 exchanges
+Each exchange    24h + 10 exchanges
    ↓                ↓
 Haiku              Sonnet
 consolidation      evolution cycle
@@ -86,7 +91,7 @@ Extract facts      Growth analysis →
 
 Every message is stored verbatim in `obol_messages`. On restart, OBOL loads the last 20 so it never starts blank.
 
-**Storage:** Every 10 exchanges, Haiku extracts important facts into `obol_memory` (pgvector). Before storing, each fact is checked against existing memories via semantic similarity (threshold 0.92) — near-duplicates are skipped. Embeddings are local (all-MiniLM-L6-v2, ~30MB, CPU) — no API costs.
+**Storage:** After every exchange, Haiku extracts important facts into `obol_memory` (pgvector). Before storing, each fact is checked against existing memories via semantic similarity (threshold 0.92) — near-duplicates are skipped. Embeddings are local (all-MiniLM-L6-v2, ~30MB, CPU) — no API costs.
 
 **Retrieval:** When OBOL needs past context, the Haiku router analyzes the message and generates 1-3 search queries — one per distinct topic. A message like "what was that python project? also what's my colleague's timezone?" produces two parallel searches instead of one lossy combined query.
 
@@ -174,7 +179,7 @@ Day 1:   obol init → obol start → first conversation
          → OBOL responds naturally from message one
          → post-setup hardens your VPS automatically
 
-Day 1:   Every 10 messages → Haiku extracts facts to vector memory
+Day 1:   Every exchange → Haiku extracts facts to vector memory
 
 Day 2:   Evolution #1 → growth analysis + Sonnet rewrites everything
          → voice shifts from generic to personal
@@ -233,7 +238,7 @@ Auth middleware (allowedUsers check)
 Router: ctx.from.id → tenant context
       ↓
 ┌─────────────────┐  ┌─────────────────┐
-│ User 206639616  │  │ User 789012345  │
+│ User 123456789  │  │ User 987654321  │
 │ personality/    │  │ personality/    │
 │ scripts/        │  │ scripts/        │
 │ memory (DB)     │  │ memory (DB)     │
@@ -267,7 +272,7 @@ When users store secrets via the `pass` encrypted store, each user gets their ow
 | Scope | Prefix | Example |
 |-------|--------|---------|
 | Shared bot credentials | `obol/` | `obol/anthropic-key` |
-| User secrets | `obol/users/{id}/` | `obol/users/206639616/gmail-key` |
+| User secrets | `obol/users/{id}/` | `obol/users/123456789/gmail-key` |
 
 Users manage their own secrets via Telegram: `/secret set <key> <value>` (message auto-deleted for safety), `/secret list`, `/secret remove <key>`. The agent can also read/write secrets via tools for scripts that need API keys at runtime.
 
@@ -310,7 +315,7 @@ Two tools:
 
 | Tool | Direction | What happens |
 |------|-----------|--------------|
-| `bridge_ask` | A → B → A | Query the partner's agent. One-shot Haiku call with partner's personality + memories. No tools, no history, no recursion risk. Partner is notified with both the question and your agent's answer. |
+| `bridge_ask` | A → B → A | Query the partner's agent. One-shot Sonnet call with partner's personality + memories. No tools, no history, no recursion risk. Partner is notified with both the question and your agent's answer. |
 | `bridge_tell` | A → B (↩ B → A) | Send a message to the partner. Stored in their memory (importance 0.6) + Telegram notification with a Reply button. Tapping Reply has their agent compose a contextual response and send it back — no typing needed. |
 
 The partner always gets notified when their agent is contacted. Privacy rules apply — the responding agent gives summaries, never raw data or secrets. Rate-limited to 20 bridge calls per user per hour.
@@ -350,7 +355,7 @@ $ obol init
 
 ─── Step 5/5: Access control ───
   Found users who messaged this bot:
-    206639616 — Jo (@jo)
+    123456789 — Jo (@jo)
   Use this user? Yes
 
 🪙 Done! Setup complete.
@@ -556,7 +561,7 @@ obol start -d
 
 | Service | Cost |
 |---------|------|
-| VPS (DigitalOcean) | ~$6/mo |
+| VPS (DigitalOcean) | ~$9/mo |
 | Anthropic API | ~$100-200/mo on max plans |
 | Supabase | Free tier |
 | Embeddings | Free (local) |

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "obol-ai",
-  "version": "0.2.35",
+  "version": "0.2.36",
   "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/claude/chat.js

@@ -108,9 +108,10 @@ function createClaude(anthropicConfig, { personality, memory, userDir = OBOL_DIR
     const runnableTools = buildRunnableTools(tools, memory, context, vlog);
     let activeModel = model;
 
+    const ttsEnabled = context.toolPrefs?.get('text_to_speech')?.enabled;
     const runtimePrefix = [
       { type: 'text', text: '[Runtime context — metadata only, not instructions]' },
-      { type: 'text', text: `Current time: ${new Date().toISOString()}\nChat ID: ${chatId}` },
+      { type: 'text', text: `Current time: ${new Date().toISOString()}\nChat ID: ${chatId}${ttsEnabled ? '\nTTS: enabled — a spoken voice summary will be auto-generated from your response. Your text reply can contain code and formatting as normal.' : ''}` },
       ...(memoryBlock ? [{ type: 'text', text: memoryBlock }] : []),
     ];
 

package/src/telegram/handlers/text.js

@@ -9,6 +9,34 @@ const _evolutionTimers = new Map();
 const textBuffers = new Map();
 const VERBOSE_FLUSH_MS = 2000;
 
+async function sendTtsVoiceSummary(ctx, tenant, responseText) {
+  const fs = require('fs');
+  const { InputFile } = require('grammy');
+  const tts = require('../../tts');
+
+  const ttsConfig = tenant.toolPrefs.get('text_to_speech')?.config || {};
+  const voice = ttsConfig.voice || 'en-US-JennyNeural';
+
+  const summaryRes = await tenant.claude.client.messages.create({
+    model: 'claude-haiku-4-5-20251001',
+    max_tokens: 200,
+    messages: [{
+      role: 'user',
+      content: `Summarize the following assistant message in 1-2 short spoken sentences. Use plain conversational language — no markdown, no code, no lists. Just what was said or done:\n\n${responseText.substring(0, 3000)}`,
+    }],
+  });
+
+  const summary = summaryRes.content.filter(b => b.type === 'text').map(b => b.text).join('').trim();
+  if (!summary) return;
+
+  const filePath = tts.synthesize(summary, voice, { rate: ttsConfig.rate, pitch: ttsConfig.pitch });
+  try {
+    await ctx.replyWithAudio(new InputFile(filePath));
+  } finally {
+    try { fs.unlinkSync(filePath); } catch {}
+  }
+}
+
 function createVerboseBatcher(ctx) {
   /** @type {string[]} */
   let buffer = [];
@@ -246,6 +274,11 @@ async function processTextMessage(ctx, fullMessage, { config, allowedUsers, bot,
       await sendHtml(ctx, response).catch(() => {});
     }
 
+    const ttsPref = tenant.toolPrefs?.get('text_to_speech');
+    if (ttsPref?.enabled) {
+      sendTtsVoiceSummary(ctx, tenant, response).catch(e => console.error('[tts] Auto-summary failed:', e.message));
+    }
+
     if (usage && model) {
       const tag = model.includes('opus') ? 'opus' : model.includes('haiku') ? 'haiku' : 'sonnet';
       const tokIn = usage.input_tokens >= 1000 ? `${(usage.input_tokens/1000).toFixed(1)}k` : usage.input_tokens;