OpenClaw News
OpenClaw News Team··13 min read·

Give OpenClaw a Voice: Building a Talking AI Agent with ElevenLabs and Telegram

Turn your OpenClaw agent into a voice-powered assistant that listens and speaks through Telegram. This step-by-step guide covers integrating ElevenLabs text-to-speech and Telegram voice messages to create a seamless, conversational AI loop you can use from anywhere.

GuideVoiceElevenLabsTelegramIntegrationTutorial
Give OpenClaw a Voice: Building a Talking AI Agent with ElevenLabs and Telegram

Text is fine. Text is functional. But when your AI agent talks back to you — in a natural, expressive human voice — something clicks. The interaction stops feeling like typing commands into a terminal and starts feeling like a conversation with a colleague who happens to have perfect memory and limitless patience.

This guide is about building exactly that: an OpenClaw agent that listens to your voice messages on Telegram and responds with natural spoken audio, powered by ElevenLabs' industry-leading text-to-speech engine. No microphone setup on your workstation. No browser tab left open. Just open Telegram on your phone, hold the record button, speak your request, and hear your agent respond in a voice so natural that your friends will think you are talking to a real person.

By the end of this article, you will have a fully operational voice-in, voice-out loop running through Telegram — accessible from any device, anywhere in the world.

Why Telegram + ElevenLabs?

Before diving into configuration, it is worth understanding why this particular combination is so powerful.

Telegram as the Interface

Telegram is arguably the best messaging platform for AI agent integration, and OpenClaw's Telegram skill takes full advantage of its capabilities:

ElevenLabs as the Voice

ElevenLabs has emerged as the clear leader in AI text-to-speech for good reason:

The combination of Telegram's ubiquitous, low-friction voice messaging and ElevenLabs' natural speech creates an experience that feels less like using a tool and more like chatting with a very capable friend.

Prerequisites

Before starting, make sure you have the following:

  1. A running OpenClaw instance — either local or on a server. If you are running OpenClaw on a Raspberry Pi or home server, that works perfectly.
  2. An ElevenLabs account — the free tier includes enough credits to test (10,000 characters/month), but the Starter plan ($5/month for 30,000 characters) is recommended for regular use.
  3. A Telegram account — you will need this to create a bot.
  4. OpenClaw's Telegram skill installed — we will cover this in the setup.

Step 1: Create Your Telegram Bot

If you have not already connected OpenClaw to Telegram, the first step is creating a bot through Telegram's BotFather:

  1. Open Telegram and search for @BotFather.
  2. Send /newbot and follow the prompts to name your bot.
  3. BotFather will give you a Bot Token — a long string that looks like 7123456789:ABCdefGHIjklMNOpqrsTUVwxyz. Save this securely.
You:       /newbot
BotFather: Alright, a new bot. How are we going to call it? 
           Please choose a name for your bot.
You:       My OpenClaw Agent
BotFather: Good. Now let's choose a username for your bot.
You:       my_openclaw_bot
BotFather: Done! Congratulations on your new bot. You can find it at 
           t.me/my_openclaw_bot. Use this token to access the HTTP API:
           7123456789:ABCdefGHIjklMNOpqrsTUVwxyz

Security Tip: Restrict your bot so only you can interact with it. BotFather's /setjoingroups command can disable group access, and OpenClaw's Telegram skill has an allowed_users setting to whitelist specific Telegram user IDs.

Step 2: Install and Configure the Telegram Skill

Install the Telegram skill from ClawHub:

openclaw skills install telegram-bot

Then configure it in your OpenClaw config:

// ~/.openclaw/openclaw.json
{
  skills: {
    telegram: {
      bot_token: "your-bot-token-from-botfather",
      allowed_users: [123456789],          // Your Telegram user ID
      auto_start: true,                    // Start the bot when OpenClaw launches

      // Voice message handling
      voice_messages: {
        enabled: true,
        transcription_provider: "whisper", // Uses your existing STT config
        respond_with_voice: true,          // Reply with voice, not text
        also_send_text: true,              // Include a text transcript too
      },
    },
  },
}

The also_send_text: true option is particularly useful — it means you get both the spoken audio response and a text version, so you can skim replies when you cannot listen (in a meeting, on a noisy train, etc.).

Find your Telegram user ID by messaging @userinfobot on Telegram. It will reply with your numeric ID.

Step 3: Set Up ElevenLabs Text-to-Speech

Sign up at elevenlabs.io and grab your API key from the Profile settings page.

Now configure ElevenLabs as your TTS provider in OpenClaw:

// ~/.openclaw/openclaw.json
{
  voice: {
    tts: {
      provider: "elevenlabs",
      api_key: "your-elevenlabs-api-key",
      voice_id: "9BWtsMINqrJLrRacOk9x",    // "Aria" — an expressive, natural female voice
      model: "eleven_turbo_v2_5",            // Lowest latency model
      stability: 0.5,                        // 0-1: lower = more expressive
      similarity_boost: 0.75,                // 0-1: higher = closer to base voice
      output_format: "mp3_44100_128",        // High quality for Telegram
    },
  },
}

Choosing the Right Voice

ElevenLabs offers a Voice Library with hundreds of options. Here are some popular choices for an AI assistant from their current default lineup:

Voice ID Description Best For
Aria 9BWtsMINqrJLrRacOk9x Expressive, natural, female General assistant
Darian (check dashboard) Warm, grounded storyteller, male Briefings & narration
Elara (check dashboard) Crisp, professional narrator, female Business use
Elowen (check dashboard) Upbeat, modern narrator, female Casual interactions
Baxter (check dashboard) Dry, calm Australian, male Relaxed assistant

Note: ElevenLabs is transitioning its default voices throughout 2026. Legacy voices like Adam, Rachel, and Antoni are being phased out by December 31, 2026. Check your ElevenLabs dashboard for the latest voice IDs — navigate to Voices > Voice Library and click any voice to copy its ID.

You can preview all voices at elevenlabs.io/voice-library before deciding.

Tuning Voice Parameters

The stability and similarity_boost settings let you fine-tune how the voice sounds:

// More expressive, conversational tone
{ stability: 0.35, similarity_boost: 0.7 }

// More consistent, professional tone
{ stability: 0.7, similarity_boost: 0.85 }

Step 4: Configure the Complete Voice Loop

Now tie it all together. The magic happens when you configure the Telegram skill to route voice messages through the full pipeline — transcription in, LLM processing, and voice out:

// ~/.openclaw/openclaw.json — complete voice configuration
{
  voice: {
    stt: {
      provider: "whisper",
      model: "whisper-1",
      api_key: "${OPENAI_API_KEY}",
    },
    tts: {
      provider: "elevenlabs",
      api_key: "${ELEVENLABS_API_KEY}",
      voice_id: "9BWtsMINqrJLrRacOk9x",
      model: "eleven_turbo_v2_5",
      stability: 0.5,
      similarity_boost: 0.75,
      output_format: "mp3_44100_128",
    },
  },
  skills: {
    telegram: {
      bot_token: "${TELEGRAM_BOT_TOKEN}",
      allowed_users: [123456789],
      auto_start: true,
      voice_messages: {
        enabled: true,
        transcription_provider: "whisper",
        respond_with_voice: true,
        also_send_text: true,
        max_voice_duration: 120,        // Max response length in seconds
        silence_trimming: true,         // Remove leading/trailing silence
      },
    },
  },
}

Restart OpenClaw to apply the configuration:

openclaw restart

Step 5: Test the Voice Loop

Open Telegram on your phone, find your bot, and send a voice message. Hold the microphone button and say something like:

"Hey, what is the weather forecast for tomorrow?"

Within a few seconds, you should receive two messages back:

  1. A voice note with the agent's response spoken in your chosen ElevenLabs voice.
  2. A text message with the transcript of the response.

If that works, congratulations — you have a talking AI agent in your pocket.

Verifying Each Stage

If something is not working, test each component individually:

# Test Telegram connection
openclaw telegram test

# Test speech-to-text
openclaw voice test-stt

# Test text-to-speech
openclaw voice test-tts

# Test the full pipeline
openclaw voice test-pipeline

Real-World Use Cases

Once the voice loop is running, the use cases multiply quickly. Here are some patterns that OpenClaw users have found particularly valuable.

The Commute Briefing

Send a voice message while walking to work: "Give me my morning briefing — calendar, emails, and anything urgent." OpenClaw gathers your calendar events, scans your inbox for flagged items, checks your task list for overdue items, and speaks a concise summary back to you. You are caught up before you reach the office.

Hands-Free Research

Cooking dinner and need to look something up: "Research the latest changes to the Australian solar rebate program and give me a summary." Your agent researches the topic, synthesizes the findings, and speaks the summary while you stir the risotto. No greasy fingerprints on your phone screen.

Voice Journaling

At the end of the day: "Save a journal entry: Today I finalized the proposal for the Henderson project. The client pushed back on the timeline so I need to revise the milestones by Friday. Also had a good idea about automating the reporting pipeline — remind me about that on Monday." OpenClaw transcribes and stores the entry, and sets a reminder for Monday about the pipeline idea.

Managing Your Team

You are running between meetings: "Send a message to the dev channel: standup is moved to 2 PM today. Also tell Sarah that I reviewed her PR and it looks good — she can merge when ready." OpenClaw sends both messages through your configured messaging integrations while you are already halfway to the next meeting room.

Quick Calculations and Lookups

Driving (safely pulled over, of course): "What is 15 percent of 847,000? And what was our revenue last quarter?" OpenClaw does the math, pulls the figure from your records, and speaks both answers.

Advanced: Custom Voice Personas

One of the more creative applications is giving different types of responses different voices. You can configure OpenClaw to use different ElevenLabs voices based on the context:

{
  voice: {
    tts: {
      provider: "elevenlabs",
      api_key: "${ELEVENLABS_API_KEY}",
      default_voice: "9BWtsMINqrJLrRacOk9x",    // Aria for general responses

      persona_voices: {
        briefing: {
          voice_id: "EIdfNdxb4fnsE39tEAB1",     // Lawrence for morning briefings
          stability: 0.7,
        },
        creative: {
          voice_id: "TDHOWxVtDS0zj6s4Jgg6",     // Alicia for brainstorming
          stability: 0.3,
        },
        urgent: {
          voice_id: "n5FCw0ouOVMADam2lTvQ",     // Caleb for urgent notifications
          stability: 0.8,
        },
      },
    },
  },
}

You can then say: "Put on your creative voice and brainstorm ten names for my new podcast." OpenClaw switches to the Alicia voice with higher expressiveness for the brainstorming session, then reverts to Aria for your next regular request.

Cost Considerations

Running a voice-enabled agent through Telegram is surprisingly affordable:

Component Cost Notes
Telegram Bot Free No cost for personal bots
Whisper API (STT) ~$0.006/min Transcribing your voice messages
ElevenLabs (TTS) ~$0.06/1K chars Speaking responses (Turbo v2.5 model)
OpenClaw Free Open source

A typical interaction — a 15-second voice message generating a 500-character response — costs roughly $0.03 per exchange for STT and TTS combined (about $0.0015 for Whisper + $0.03 for ElevenLabs Turbo). If you send 20 voice messages per day, that is under $0.60/day or about $18/month. Cheaper than most SaaS subscriptions and significantly more useful.

The ElevenLabs Starter plan ($5/month) gives you 30,000 characters, which covers roughly 60 average-length responses. The Creator plan ($22/month) bumps this to 100,000 characters — enough for heavy daily use.

Reducing Costs

If you want to cut costs further:

{
  skills: {
    telegram: {
      voice_messages: {
        max_tts_characters: 800,          // Text-only if response exceeds 800 chars
        voice_summary_for_long: true,     // Speak a summary, send full text separately
      },
    },
  },
}

Troubleshooting

Voice messages are not being received

Ensure your bot token is correct and that the Telegram skill is running (openclaw status). Check that your Telegram user ID is in the allowed_users list.

Transcription is inaccurate

Background noise is the most common culprit. Telegram compresses voice messages heavily, which can reduce transcription quality. Speaking clearly and minimizing background noise helps significantly.

ElevenLabs responses sound truncated

Check the max_voice_duration setting. If your response is being cut off, increase this value. Also verify that your ElevenLabs plan has sufficient character quota remaining.

Latency is too high

The Turbo v2.5 model is the fastest ElevenLabs option. If latency is still an issue, check your internet connection speed and consider running OpenClaw on a server with a fast connection rather than a local machine behind a slow upload link.

Bot is not responding

Run openclaw logs --skill telegram to see the Telegram skill's log output. Common issues include expired bot tokens and misconfigured webhook URLs.

Conclusion

Giving OpenClaw a voice through Telegram and ElevenLabs transforms an already capable AI agent into something that feels genuinely futuristic. You pull out your phone, hold the microphone button, and have a natural conversation with an AI that knows your preferences, remembers your projects, and can actually do things — send messages, research topics, manage your schedule, and control your tools.

The setup takes about 20 minutes. The cost is well under a dollar a day for regular use. And the experience is unlike anything else available in the open-source AI space.

Your agent is not just smart anymore. It has a voice. And it is waiting for you on Telegram.

Share this article

Recommended Reading