obol-ai 0.2.16 → 0.2.18

package/CHANGELOG.md

@@ -1,3 +1,14 @@
+## 0.2.18
+- remove evolution progress bar from status UI
+- bidirectional bridge with reply button + memory_remove tool
+- update background tasks section in readme
+- add status UI screenshot to readme
+- update readme with stop controls, commands, and model escalation
+
+## 0.2.17
+- add force stop button to instantly abort mid-tool execution
+- replace web_fetch with native web_search tool
+
 ## 0.2.16
 - add chat_history tool for retrieving past conversations by date
 - add stop button to status UI with concurrent update processing

package/README.md

@@ -22,7 +22,7 @@ obol start -d   # runs as background daemon (auto-installs pm2)
 
 🧠 **Living memory** — Vector memory with semantic search. Haiku routes queries and rewrites them for better embedding hits. Free local embeddings.
 
-🤖 **Smart routing** — Haiku decides per-message: does it need memory? Sonnet or Opus? No wasted API calls
+🤖 **Smart routing** — Haiku decides per-message: does it need memory? Sonnet or Opus? Auto-escalates to Sonnet when tool use is needed. No wasted API calls
 
 💰 **Prompt caching** — Static system prompt and conversation history prefix are cached via Anthropic's prompt caching, cutting ~85% of repeated input token costs across turns
 
@@ -59,9 +59,9 @@ User message
     ↓             ↓
 Memory recall   Model selection
     ↓             ↓
-Multi-query     Sonnet (default)
-ranked recall   or Opus (complex)
-    ↓             ↓
+Multi-query     Haiku → Sonnet (auto-
+ranked recall   escalates on tool use)
+    ↓             or Opus (complex)
     └──────┬──────┘
            ↓
    Claude (tool use loop)
@@ -199,21 +199,32 @@ Month 6: evolution/ has 180+ archived souls
 
 ### Background Tasks
 
-Heavy work runs in the background. The main conversation stays responsive.
+Heavy work runs in the background with its own live status UI. The main conversation stays responsive — you can keep chatting while tasks run.
 
 ```
 You: "research the best coworking spaces in Barcelona"
-OBOL: "On it 🪙"
-
-[30s] ⏳ Found 15 spaces, filtering by reviews...
-[60s] ⏳ Narrowed to top 7, checking prices...
+OBOL: spawns BG #1 with live status
 
 You: "what time is it?"
 OBOL: "11:42 PM CET"
 
-[90s] ✅ Done! Here are the top 5 coworking spaces: ...
+✅ BG #1 done (1m 32s)
+Here are the top 5 coworking spaces: ...
 ```
 
+### Live Status & Stop Controls
+
+![Status UI](docs/obol-status.png)
+
+Every request shows a live status message with elapsed time, model routing info, and what tools are being used. Two inline buttons let you cancel:
+
+| Button | Behavior |
+|--------|----------|
+| **■ Stop** | Cancels after the current API call finishes |
+| **■ Force Stop** | Instantly aborts mid-tool — races the handler and returns immediately |
+
+The `/stop` command also works as a text alternative.
+
 ## Multi-User Architecture
 
 One Telegram bot token, one Node.js process, full per-user isolation.
@@ -275,29 +286,39 @@ Each new user starts fresh. Their bot evolves independently from every other use
 
 ### Bridge (couples / roommates / teams)
 
-When two users share the same OBOL instance, their agents can talk to each other.
+When two users share the same OBOL instance, their agents can talk to each other — bidirectionally.
 
 ```
 User A: "what does Jo want for dinner tonight?"
 Agent A: → bridge_ask → Agent B (one-shot, no tools, no history)
 Agent B: "Jo mentioned craving Thai food earlier today"
 Agent A: "Jo's been wanting Thai — maybe suggest pad see ew?"
+
+Jo gets: "🪙 Your partner's agent asked: 'what does Jo want for dinner?'
+          Your agent answered: 'Jo mentioned craving Thai food earlier today'"
 ```
 
 ```
 User A: "remind Jo I'll be home late"
 Agent A: → bridge_tell → stores in Agent B's memory + Telegram notification
-Jo gets: "🪙 Message from your partner's agent: I'll be home late"
+
+Jo gets: "🪙 Message from your partner's agent:
+          'I'll be home late'"
+          [↩ Reply]
+
+Jo taps Reply → Jo's agent reads recent bridge context, composes a reply
+             → sends back via bridge_tell
+A gets: "🪙 Message from your partner's agent: 'Got it, I'll start dinner around 7'"
 ```
 
 Two tools:
 
 | Tool | Direction | What happens |
 |------|-----------|--------------|
-| `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. |
-| `bridge_tell` | A → B | Send a message to the partner. Stored in their memory (importance 0.6) + Telegram notification. Their agent picks it up as context in future conversations. |
+| `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_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.
+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.
 
 Enable during `obol init` (auto-prompted when 2+ users are added) or toggle later with `obol config` → Bridge.
 
@@ -471,6 +492,7 @@ Or edit `~/.obol/config.json` directly:
 /memory     — Search or view memory stats
 /recent     — Last 10 memories
 /today      — Today's memories
+/events     — Show upcoming scheduled events
 /tasks      — Running background tasks
 /status     — Bot status, uptime, evolution progress, traits
 /backup     — Trigger GitHub backup
@@ -478,6 +500,11 @@ Or edit `~/.obol/config.json` directly:
 /traits     — View or adjust personality traits (0-100)
 /secret     — Manage per-user encrypted secrets
 /evolution  — Evolution progress
+/verbose    — Toggle verbose mode on/off
+/toolimit   — View or set max tool iterations per message
+/tools      — Toggle optional tools on/off
+/stop       — Stop the current request
+/upgrade    — Check for updates and upgrade
 /help       — Show available commands
 ```
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "obol-ai",
-  "version": "0.2.16",
+  "version": "0.2.18",
   "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/bridge.js

@@ -106,7 +106,8 @@ async function bridgeAsk(question, fromUserId, config, notifyFn, targetId) {
 
   if (notifyFn) {
     try {
-      await notifyFn(partnerUserId, `🪙 Your partner's agent asked about you:\n"${question}"`);
+      const preview = answer.length > 200 ? `${answer.substring(0, 200)}…` : answer;
+      await notifyFn(partnerUserId, `🪙 Your partner's agent asked: "${question}"\nYour agent answered: "${preview}"`);
     } catch (e) {
       console.error(`[bridge] Notify failed for ${partnerUserId}:`, e.message);
     }
@@ -170,7 +171,8 @@ async function bridgeTell(message, fromUserId, config, notifyFn, targetId) {
 
   if (notifyFn) {
     try {
-      await notifyFn(partnerUserId, `🪙 Message from your partner's agent:\n"${message}"`);
+      const replyMarkup = { inline_keyboard: [[{ text: '↩ Reply', callback_data: `bridge:reply:${fromUserId}` }]] };
+      await notifyFn(partnerUserId, `🪙 Message from your partner's agent:\n"${message}"`, { reply_markup: replyMarkup });
     } catch (e) {
       console.error(`[bridge] Notify failed for ${partnerUserId}:`, e.message);
     }

package/src/claude/chat.js

@@ -16,6 +16,7 @@ function createClaude(anthropicConfig, { personality, memory, userDir = OBOL_DIR
   const histories = new ChatHistory(50);
   const chatLocks = new Map();
   const chatAbortControllers = new Map();
+  const chatForceControllers = new Map();
 
   const tools = buildTools(memory, { bridgeEnabled });
 
@@ -45,7 +46,9 @@ function createClaude(anthropicConfig, { personality, memory, userDir = OBOL_DIR
 
     const releaseLock = await acquireChatLock(chatId);
     const abortController = new AbortController();
+    const forceController = new AbortController();
     chatAbortControllers.set(chatId, abortController);
+    chatForceControllers.set(chatId, forceController);
 
     const history = histories.get(chatId);
 
@@ -95,6 +98,7 @@ function createClaude(anthropicConfig, { personality, memory, userDir = OBOL_DIR
     ];
     context._reloadPersonality = reloadPersonality;
     context._abortSignal = abortController.signal;
+    context._forceSignal = forceController.signal;
     const runnableTools = buildRunnableTools(tools, memory, context, vlog);
     let activeModel = model;
 
@@ -198,6 +202,7 @@ function createClaude(anthropicConfig, { personality, memory, userDir = OBOL_DIR
       throw e;
     } finally {
       chatAbortControllers.delete(chatId);
+      chatForceControllers.delete(chatId);
       releaseLock();
     }
   }
@@ -211,6 +216,14 @@ function createClaude(anthropicConfig, { personality, memory, userDir = OBOL_DIR
     return false;
   }
 
+  function forceStopChat(chatId) {
+    const controller = chatAbortControllers.get(chatId);
+    const force = chatForceControllers.get(chatId);
+    if (force) force.abort();
+    if (controller) { controller.abort(); return true; }
+    return false;
+  }
+
   function reloadPersonality() {
     const pDir = userDir ? path.join(userDir, 'personality') : undefined;
     const newPersonality = require('../personality').loadPersonality(pDir);
@@ -236,7 +249,7 @@ function createClaude(anthropicConfig, { personality, memory, userDir = OBOL_DIR
     return histories.estimateTokens(id, baseSystemPrompt.length);
   }
 
-  return { chat, client, reloadPersonality, clearHistory, injectHistory, getContextStats, stopChat };
+  return { chat, client, reloadPersonality, clearHistory, injectHistory, getContextStats, stopChat, forceStopChat };
 }
 
 module.exports = { createClaude };

package/src/claude/prompt.js

@@ -114,8 +114,8 @@ Categories: \`fact\`, \`preference\`, \`decision\`, \`lesson\`, \`person\`, \`pr
 Read and write files within your workspace. Parent directories created automatically.
 Cannot access paths outside workspace or /tmp.
 
-### Web (\`web_fetch\`)
-Fetch and extract readable content from any URL via Jina reader.
+### Web (\`web_search\`)
+Search the web for current information.
 
 ### Vercel (\`vercel_deploy\`, \`vercel_list\`)
 Deploy directories to Vercel. Ship websites, dashboards, web apps.
@@ -252,7 +252,7 @@ Structure tips:
 - Search memory before claiming you don't know something
 - Use \`store_secret\`/\`read_secret\` for all credential operations
 - If a user sends what appears to be an API key, token, or credential in conversation, immediately warn them that it's visible in chat history, tell them to revoke/rotate it, and direct them to use \`/secret set <key> <value>\` instead
-- After executing tools (exec, web_fetch, read_secret, etc.), ALWAYS provide a text response summarizing what you found or did. Never end your turn with only tool calls and no text reply — the user cannot see tool results directly, they only see your text responses
+- After executing tools (exec, web_search, read_secret, etc.), ALWAYS provide a text response summarizing what you found or did. Never end your turn with only tool calls and no text reply — the user cannot see tool results directly, they only see your text responses
 `);
 
   return parts.join('\n');

package/src/claude/tool-registry.js

@@ -32,8 +32,9 @@ const INPUT_SUMMARIES = {
   read_file: (i) => i.path,
   memory_search: (i) => i.query,
   memory_add: (i) => `[${i.category || 'fact'}]`,
+  memory_remove: (i) => i.ids?.join(', '),
   memory_query: (i) => `${i.date || ''}${i.tags ? ' #' + i.tags.join(' #') : ''}${i.category ? ' [' + i.category + ']' : ''}`.trim() || 'all',
-  web_fetch: (i) => i.url,
+  web_search: (i) => i.query,
   background_task: (i) => i.task?.substring(0, 60),
   schedule_event: (i) => `${i.title} @ ${i.due_at}${i.cron_expr ? ` [${i.cron_expr}]` : ''}`,
   cancel_event: (i) => i.event_id,
@@ -95,7 +96,9 @@ function buildRunnableTools(tools, memory, context, vlog) {
     .map(tool => ({
       ...tool,
       run: async (input) => {
-        if (context._abortSignal?.aborted) return 'Aborted.';
+        const signal = context._abortSignal;
+        const forceSignal = context._forceSignal;
+        if (signal?.aborted) return 'Aborted.';
         const inputSummary = summarizeInput(tool.name, input);
         vlog(`[tool] ${tool.name}: ${inputSummary}`);
         context._onToolStart?.(tool.name, inputSummary);
@@ -104,8 +107,14 @@ function buildRunnableTools(tools, memory, context, vlog) {
         if (!handler) return `Unknown tool: ${tool.name}`;
 
         try {
-          return await handler(input, memory, context);
+          if (!forceSignal) return await handler(input, memory, context);
+          if (forceSignal.aborted) return 'Aborted.';
+          const forcePromise = new Promise((resolve) => {
+            forceSignal.addEventListener('abort', () => resolve('Aborted.'), { once: true });
+          });
+          return await Promise.race([handler(input, memory, context), forcePromise]);
         } catch (e) {
+          if (signal?.aborted || forceSignal?.aborted) return 'Aborted.';
           return `Error: ${e.message}`;
         }
       },

package/src/claude/tools/memory.js

@@ -27,6 +27,17 @@ const definitions = [
       required: ['content'],
     },
   },
+  {
+    name: 'memory_remove',
+    description: 'Remove one or more memories by their IDs. Use memory_query or memory_search first to get IDs.',
+    input_schema: {
+      type: 'object',
+      properties: {
+        ids: { type: 'array', items: { type: 'string' }, description: 'Memory IDs to remove' },
+      },
+      required: ['ids'],
+    },
+  },
   {
     name: 'memory_query',
     description: 'Filter memories by tag, date, category, source, or importance. Use for "what did we do today", "anything tagged X", "all decisions this week".',
@@ -47,7 +58,7 @@ const definitions = [
 function formatMemory(m) {
   const date = m.created_at ? new Date(m.created_at).toISOString().slice(0, 10) : null;
   const tags = m.tags?.length ? ` #${m.tags.join(' #')}` : '';
-  return `[${date || '?'}] [${m.category}] ${m.content}${tags}`;
+  return `[id:${m.id}] [${date || '?'}] [${m.category}] ${m.content}${tags}`;
 }
 
 const handlers = {
@@ -69,6 +80,11 @@ const handlers = {
     return `Stored memory: ${result.id}`;
   },
 
+  async memory_remove(input, memory) {
+    await Promise.all(input.ids.map(id => memory.forget(id)));
+    return `Removed ${input.ids.length} memory${input.ids.length !== 1 ? 'ies' : ''}`;
+  },
+
   async memory_query(input, memory) {
     const results = await memory.query({
       date: input.date,

package/src/claude/tools/web.js

@@ -1,28 +1,8 @@
-const { isAllowedUrl } = require('../../sanitize');
-
 const definitions = [{
-  name: 'web_fetch',
-  description: 'Fetch and extract readable content from a URL.',
-  input_schema: {
-    type: 'object',
-    properties: {
-      url: { type: 'string', description: 'URL to fetch' },
-    },
-    required: ['url'],
-  },
+  type: 'web_search_20250305',
+  name: 'web_search',
 }];
 
-const handlers = {
-  async web_fetch(input) {
-    if (!isAllowedUrl(input.url)) return 'Blocked: URL points to a private/internal address.';
-    const jinaUrl = `https://r.jina.ai/${input.url}`;
-    const res = await fetch(jinaUrl, {
-      headers: { 'Accept': 'text/markdown' },
-    });
-    if (!res.ok) return `Failed to fetch: HTTP ${res.status}`;
-    const text = await res.text();
-    return text.substring(0, 15000);
-  },
-};
+const handlers = {};
 
 module.exports = { definitions, handlers };

package/src/evolve/prompts.js

@@ -17,7 +17,7 @@ Third person factual profile: name, location, timezone, nationality, job, skills
 Operational manual written as instructions to yourself. Focus on owner-specific workflows, service integrations, and lessons learned from conversations.
 
 **Do NOT include in AGENTS.md** — these are already hardcoded in the base system prompt and must not be duplicated:
-- Tool documentation (exec, memory_*, read_file, write_file, web_fetch, vercel_*, background_task, store_secret, read_secret, list_secrets, send_file, telegram_ask, bridge_*)
+- Tool documentation (exec, memory_*, read_file, write_file, web_search, vercel_*, background_task, store_secret, read_secret, list_secrets, send_file, telegram_ask, bridge_*)
 - Telegram Formatting rules
 - Safety Rules (Never/Always)
 - Workspace Structure

package/src/telegram/bot.js

@@ -48,7 +48,7 @@ function createBot(telegramConfig, config) {
   }
 
   bot.use(sequentialize((ctx) => {
-    if (ctx.callbackQuery?.data?.startsWith('stop:')) return undefined;
+    if (ctx.callbackQuery?.data?.startsWith('stop:') || ctx.callbackQuery?.data?.startsWith('force:')) return undefined;
     return ctx.chat?.id.toString();
   }));
 

package/src/telegram/commands/status.js

@@ -1,8 +1,7 @@
 const path = require('path');
 const { getTenant } = require('../../tenant');
-const { loadConfig } = require('../../config');
 const { loadTraits } = require('../../personality');
-const { evolve, loadEvolutionState } = require('../../evolve');
+const { loadEvolutionState } = require('../../evolve');
 const { getMaxToolIterations } = require('../../claude');
 const { termBar, formatTraits } = require('../utils');
 const { TERM_SEP } = require('../constants');
@@ -44,16 +43,13 @@ function register(bot, config) {
     );
 
     const evoState = loadEvolutionState(tenant.userDir);
-    const cfg = loadConfig();
-    const intervalHours = cfg?.evolution?.intervalHours ?? 24;
-    const elapsed = evoState.lastEvolution ? (Date.now() - new Date(evoState.lastEvolution).getTime()) / 3600000 : Infinity;
-    const evoPct = Math.min(100, Math.round((elapsed / intervalHours) * 100));
-    const timeLeft = Math.max(0, intervalHours - elapsed);
     lines.push(
       ``, `EVOLUTION`,
-      `  ${termBar(evoPct)} ${evoPct}%`,
-      `  ${timeLeft < 1 ? 'ready' : `${timeLeft.toFixed(1)}h remaining`} ▪ ${evoState.evolutionCount || 0} completed`,
+      `  ${evoState.evolutionCount || 0} completed`,
     );
+    if (evoState.lastEvolution) {
+      lines.push(`  last ${new Date(evoState.lastEvolution).toLocaleDateString()}`);
+    }
 
     const personalityDir = path.join(tenant.userDir, 'personality');
     const traits = loadTraits(personalityDir);
@@ -68,18 +64,11 @@ function register(bot, config) {
     if (!ctx.from) return;
     const tenant = await getTenant(ctx.from.id, config);
     const state = loadEvolutionState(tenant.userDir);
-    const cfg = loadConfig();
-    const intervalHours = cfg?.evolution?.intervalHours ?? 24;
-    const elapsed = state.lastEvolution ? (Date.now() - new Date(state.lastEvolution).getTime()) / 3600000 : Infinity;
-    const pct = Math.min(100, Math.round((elapsed / intervalHours) * 100));
-    const timeLeft = Math.max(0, intervalHours - elapsed);
 
     const lines = [
       `◈ OBOL EVOLUTION CYCLE`,
       TERM_SEP,
       ``,
-      `  ${termBar(pct)} ${pct}%`,
-      `  ${timeLeft < 1 ? 'ready' : `${timeLeft.toFixed(1)}h remaining`}`,
       `  ${state.evolutionCount || 0} completed`,
     ];
     if (state.lastEvolution) {

package/src/telegram/handlers/callbacks.js

@@ -12,7 +12,16 @@ function registerCallbackHandler(bot, { config, pendingAsks, getTenant }) {
       const userId = ctx.from.id;
       const tenant = await getTenant(userId, config);
       const stopped = tenant?.claude?.stopChat(chatId);
-      await answer({ text: stopped ? 'Stopping' : 'Nothing to stop' });
+      await answer({ text: stopped ? 'Stopping...' : 'Nothing to stop' });
+      return;
+    }
+
+    if (data.startsWith('force:')) {
+      const chatId = parseInt(data.slice(6));
+      const userId = ctx.from.id;
+      const tenant = await getTenant(userId, config);
+      const stopped = tenant?.claude?.forceStopChat(chatId);
+      await answer({ text: stopped ? 'Force stopped' : 'Nothing to stop' });
       return;
     }
 
@@ -27,6 +36,66 @@ function registerCallbackHandler(bot, { config, pendingAsks, getTenant }) {
       return;
     }
 
+    if (data.startsWith('bridge:reply:')) {
+      const targetUserId = parseInt(data.split(':')[2]);
+      const reactingUserId = ctx.from.id;
+
+      const tenant = await getTenant(reactingUserId, config);
+      if (!tenant) return answer({ text: 'Could not load your agent' });
+
+      const { checkBridgeRateLimit, bridgeTell } = require('../../bridge');
+      const rateErr = checkBridgeRateLimit(reactingUserId);
+      if (rateErr) return answer({ text: rateErr });
+
+      let memoryContext = '';
+      if (tenant.memory) {
+        try {
+          const memories = await tenant.memory.search('message from partner bridge', { limit: 5, threshold: 0.3 });
+          if (memories.length > 0) {
+            memoryContext = '\n\n[Recent bridge messages]\n' + memories.map(m => `- ${m.content}`).join('\n');
+          }
+        } catch {}
+      }
+
+      const systemParts = [
+        'Compose a brief, natural reply to send back to your partner\'s agent via bridge. 1-3 sentences. Be genuine and respond to the most recent message from them.',
+      ];
+      if (tenant.personality?.soul) systemParts.push(`\n## Your Personality\n${tenant.personality.soul}`);
+      if (tenant.personality?.user) systemParts.push(`\n## About You\n${tenant.personality.user}`);
+      if (memoryContext) systemParts.push(memoryContext);
+
+      let replyText;
+      try {
+        const response = await tenant.claude.client.messages.create({
+          model: 'claude-haiku-4-5-20251001',
+          max_tokens: 256,
+          system: systemParts.join('\n'),
+          messages: [{ role: 'user', content: 'Compose your reply to send via bridge.' }],
+        });
+        replyText = response.content.filter(b => b.type === 'text').map(b => b.text).join('\n').trim();
+      } catch (e) {
+        console.error('[bridge:reply] Generation failed:', e.message);
+        return answer({ text: 'Failed to generate reply' });
+      }
+
+      if (!replyText) return answer({ text: 'Could not generate a reply' });
+
+      const notifyFn = (uid, msg, opts = {}) => ctx.api.sendMessage(uid, msg, opts);
+      try {
+        await bridgeTell(replyText, reactingUserId, config, notifyFn, targetUserId);
+      } catch (e) {
+        console.error('[bridge:reply] bridgeTell failed:', e.message);
+        return answer({ text: 'Failed to send reply' });
+      }
+
+      ctx.editMessageText(
+        ctx.callbackQuery.message.text + '\n\n✓ Reply sent',
+        { reply_markup: { inline_keyboard: [] } }
+      ).catch(() => {});
+
+      return answer({ text: 'Reply sent!' });
+    }
+
     if (!data.startsWith('ask:')) return answer();
     const parts = data.split(':');
     const askId = parseInt(parts[1]);

package/src/telegram/handlers/text.js

@@ -27,9 +27,9 @@ function createChatContext(ctx, tenant, config, { allowedUsers, bot, createAsk }
       sendHtml(ctx, `\`${msg}\``).catch(() => {});
     } : undefined,
     telegramAsk: (message, options, timeout) => createAsk(ctx, message, options, timeout),
-    _notifyFn: (targetUserId, message) => {
+    _notifyFn: (targetUserId, message, opts = {}) => {
       if (!allowedUsers.has(targetUserId)) throw new Error('Cannot notify user outside allowed list');
-      return bot.api.sendMessage(targetUserId, message);
+      return bot.api.sendMessage(targetUserId, message, opts);
     },
   };
 }
@@ -40,7 +40,9 @@ function createStatusTracker(ctx) {
   let statusTimer = null;
   let statusStart = null;
   let routeInfo = null;
-  const stopBtn = new InlineKeyboard().text('■ Stop', `stop:${ctx.chat.id}`);
+  const stopBtn = new InlineKeyboard()
+    .text('■ Stop', `stop:${ctx.chat.id}`)
+    .text('■ Force Stop', `force:${ctx.chat.id}`);
 
   const clear = () => {
     if (statusTimer) { clearInterval(statusTimer); statusTimer = null; }