Skip to main content

TalkiTo

GitHub Stars GitHub Forks License Discord

TalkiTo lets developers talk, slack and whatsapp with Claude Code, OpenAI Codex, Gemini CLI, OpenCode and pi. It can be used as a command-line tool, a web extension, and as a Python library.

🚀 Quick Install

curl -sSL https://raw.githubusercontent.com/robdmac/talkito/main/install.sh | bash

Option 2: PyPI

pip install talkito

Then just run:

talkito claude

Install for End Users

From Source (Stable)

# Clone the repository
git clone https://github.com/robdmac/talkito.git
cd talkito

# Create and activate virtual environment (recommended)
python3 -m venv venv
source venv/bin/activate

# Install system dependencies (macOS)
brew install portaudio

# Install package (normal install - gets updates via git pull)
pip install .

# Run this in a directory you want to use claude with
talkito claude

Install for Developers

Editable Install (Development)

# Clone the repository
git clone https://github.com/robdmac/talkito.git
cd talkito

# Create and activate virtual environment (recommended)
python3 -m venv venv
source venv/bin/activate

# Install system dependencies (macOS)
brew install portaudio

# Install in development mode (editable install)
pip install -e .

# Run this in a directory you want to use claude with
talkito claude

or for the web extension run as

talkito --mcp-sse-server

then go to chrome://extensions/ and load unpacked the extensions/chrome/ dir

Demo Video

TalkiTo Demo

AI Assistant Compatibility

AI Assistant Method Status
Claude Code Terminal Fully Supported
Codex Cli Terminal Fully Supported
Gemini CLI Terminal Fully Supported
OpenCode Terminal Fully Supported
pi Terminal Fully Supported
bolt.new Web Extension Output Only
v0.dev Web Extension Output Only
replit.com Web Extension Output Only
Other agents Terminal In Progress

Run with Claude Code

run talkito claude

Run with Codex Cli

run talkito codex

Run with Gemini CLI, OpenCode or pi

run talkito gemini, talkito opencode or talkito pi

Talk to your agent

TalkiTo listens while you hold the backtick key (`). What it hears appears in grey after your cursor as you speak and becomes the final text when you let go, which sends it. In a long hold, what you said up to each pause is typed in after about six seconds while the rest stays grey. Say "execute" to send straight away.

Use --asr-mode auto-input to listen whenever the agent is waiting for you, or --asr-mode off for speech output only. Anything the microphone picks up of TalkiTo's own voice while you hold the key is ignored.

Run as an MCP server

run talkito --mcp-server

Run the TalkiTo configuration menu

run talkito

Advanced Options

# Disable auto-skip to newer content (auto-skip is on by default)
talkito --dont-auto-skip-tts claude

# Use different TTS providers
talkito --tts-provider polly --tts-voice Matthew --tts-region us-west-2 echo "Hello with AWS"
talkito --tts-provider azure --tts-voice en-US-JennyNeural echo "Hello with Azure"
talkito --tts-provider gcloud --tts-voice en-US-Journey-F echo "Hello with Google"
talkito --tts-provider kittentts --tts-voice expr-voice-3-f echo "Hello with KittenTTS"
talkito --tts-provider kokoro --tts-voice af_heart echo "Hello with Kokoro (local)"
talkito --tts-provider piper --tts-voice en_US-lessac-medium echo "Hello with Piper (local)"
talkito --tts-provider neutts2e echo "Hello with NeuTTS (local)"

# Use different ASR providers
talkito --asr-provider gcloud --asr-language en-US claude
AZURE_SPEECH_KEY=... AZURE_SPEECH_REGION=eastus talkito --asr-provider azure claude
WHISPER_MODEL=small WHISPER_COMPUTE_TYPE=int8 talkito --asr-provider local_whisper claude
talkito --asr-language es-ES echo "Hola mundo"  # Spanish recognition

# Enable remote communication (configure via environment variables)
talkito --slack-channel '#alerts' python manage.py runserver
talkito --whatsapp-recipients +1234567890 long-running-command
talkito --sms-recipients +1234567890,+0987654321 server-monitor.sh

Using tts.py (Standalone TTS)

The TTS module can be used independently for text-to-speech operations:

#!/usr/bin/env python3
import tts

# Initialize TTS
engine = tts.detect_tts_engine()
tts.start_tts_worker(engine)

# Speak text
tts.queue_for_speech("Hello from the TTS module!")

# Wait and cleanup
import time
time.sleep(2)
tts.shutdown_tts()

Using asr.py (Standalone ASR)

The ASR module can be used independently for speech recognition:

#!/usr/bin/env python3
import asr

# Define callback for recognized text
def handle_text(text):
    print(f"You said: {text}")

# Start dictation
asr.start_dictation(handle_text)

# Keep running (press Ctrl+C to stop)
try:
    import time
    while True:
        time.sleep(1)
except KeyboardInterrupt:
    asr.stop_dictation()

Provider Configuration

Text-to-Speech (TTS) Providers

System TTS (Default)

  • macOS: Uses built-in say command
  • Linux: Uses espeak, festival, or flite (install via package manager)
  • Setup: No API key needed

OpenAI TTS

AWS Polly

Azure Speech Services

Google Cloud Text-to-Speech

ElevenLabs

  • Get API Key: https://elevenlabs.io/
  • Setup: Set ELEVENLABS_API_KEY
  • Voices: Various voice IDs available
  • Usage: Configure in code or .env file

Deepgram

  • Get API Key: https://deepgram.com/
  • Setup: Set DEEPGRAM_API_KEY
  • Voices: aura-asteria-en, aura-luna-en, aura-stella-en, and more
  • Usage: --tts-provider deepgram --tts-voice aura-asteria-en

KittenTTS (Local / Offline)

  • Install: pip install https://github.com/KittenML/KittenTTS/releases/download/0.1/kittentts-0.1.0-py3-none-any.whl soundfile phonemizer
  • Setup: No API key required. First run prompts to download the selected model (default kitten-tts-nano-0.2) into the Hugging Face cache. Configure KITTENTTS_MODEL and KITTENTTS_VOICE to pick different quality/voice options.
  • Best for: Ultra-lightweight CPU-only voices that stay on-device.
  • Usage: KITTENTTS_MODEL=kitten-tts-nano-0.2 talkito --tts-provider kittentts --tts-voice expr-voice-3-f

Kokoro (Local / Offline)

  • Install: pip install 'kokoro>=0.9.4' soundfile phonemizer
  • Setup: No API key required. TalkiTo will download Kokoro weights the first time you run it (set KOKORO_LANGUAGE, KOKORO_VOICE, KOKORO_SPEED to control defaults).
  • Best for: High-quality multilingual voices without sending audio to a cloud provider.
  • Usage: talkito --tts-provider kokoro --tts-voice af_heart --tts-language en-US

Piper (Local / Offline)

  • Install: pip install piper-tts
  • Setup: No API key required. A voice is a single ONNX file of a few tens of megabytes, downloaded on first use into ~/.cache/talkito/piper (set PIPER_VOICE and PIPER_VOICE_DIR to change the default voice or location).
  • Best for: The lightest offline option by a wide margin, and the fastest. It needs no torch at all, where the other local engines pull in hundreds of megabytes of runtime.
  • Voices: en_US-lessac-medium (default), en_GB-alan-medium, en_US-amy-medium and many more — the menu lists a shortlist, but any name Piper publishes is accepted. Full list: python -m piper.download_voices --help
  • Usage: talkito --tts-provider piper --tts-voice en_GB-alan-medium

NeuTTS 2E (Local / Offline)

  • Install: pip install neutts llama-cpp-python soundfile
  • Setup: No API key required. The first run downloads a quantized backbone and a decoder-only ONNX codec, around 600 MB together. Set NEUTTS2E_VOICE and NEUTTS2E_EMOTION to change the speaker and delivery.
  • Best for: Expressive, emotion-conditioned speech that stays on-device. It is a language model over a neural audio codec rather than a feed-forward synthesiser, so it sounds markedly less robotic than its size suggests.
  • Speakers: emily (default), paul, sophie, steven. Emotions: neutral, happy, sad, angry, surprised, fearful, disgusted.
  • Devices: On Apple Silicon the backbone runs on Metal automatically and falls back to CPU if unavailable. NEUTTS_DEVICE=cpu pins everything to CPU; NEUTTS_BACKBONE_DEVICE overrides the backbone alone.
  • Usage: NEUTTS2E_EMOTION=happy talkito --tts-provider neutts2e --tts-voice sophie

Automatic Speech Recognition (ASR) Providers

Google Speech Recognition (Default)

  • Free: No API key required
  • Limitations: Best for short utterances, requires internet
  • Usage: Default when no provider specified

Google Cloud Speech-to-Text

AssemblyAI

  • Get API Key: https://www.assemblyai.com/
  • Setup: Set ASSEMBLYAI_API_KEY
  • Features: Real-time transcription, speaker detection
  • Usage: Configure in code or .env file

Deepgram

  • Get API Key: https://deepgram.com/
  • Setup: Set DEEPGRAM_API_KEY
  • Features: Fast, accurate real-time transcription
  • Usage: Configure in code or .env file

Houndify

  • Get Credentials: https://www.houndify.com/
  • Setup: Set HOUNDIFY_CLIENT_ID and HOUNDIFY_CLIENT_KEY
  • Features: Natural language understanding
  • Usage: --asr-provider houndify

AWS Transcribe

Azure Speech Services

Local Whisper (On-Device)

  • Install: pip install faster-whisper (default) or WHISPER_COREML=1 pip install pywhispercpp for Apple Silicon/CoreML acceleration
  • Setup: No API key required. Configure WHISPER_MODEL (e.g., small, medium), WHISPER_DEVICE (cpu, cuda, or mps), and WHISPER_COMPUTE_TYPE (int8, int8_float16, etc.). Models are cached locally and TalkiTo will prompt before downloading unless TALKITO_AUTO_APPROVE_DOWNLOADS=1.
  • Usage: WHISPER_MODEL=small WHISPER_COMPUTE_TYPE=int8 talkito --asr-provider local_whisper
  • Vocabulary: Whisper is primed to expect coding terms such as "git branch" and "pull request". Add your own with TALKITO_ASR_TERMS="Kokoro, pywhispercpp".

Communication Providers (Remote Interaction)

Twilio SMS

  • Get Account: https://www.twilio.com/try-twilio
  • Setup: Set TWILIO_ACCOUNT_SID, TWILIO_AUTH_TOKEN, TWILIO_PHONE_NUMBER you will need to a verified number to avoid being filtered.
  • Features: Send command output via SMS, receive input via SMS
  • Usage: --sms-recipients +1234567890

Twilio WhatsApp

  • Get Started: https://www.twilio.com/whatsapp
  • Setup Instructions: Run talkito --setup-whatsapp for detailed setup guide
  • Required Environment Variables:
    • TWILIO_ACCOUNT_SID: Your Twilio account SID
    • TWILIO_AUTH_TOKEN: Your Twilio auth token
    • TWILIO_WHATSAPP_NUMBER: Twilio's WhatsApp number (usually +14155238886)
    • WHATSAPP_RECIPIENTS: Your WhatsApp number
    • ZROK_RESERVED_TOKEN: Your zrok reserved share token
  • Quick Setup:
    • Join Twilio WhatsApp Sandbox at https://www.twilio.com/console/sms/whatsapp/sandbox
    • Send the join code via WhatsApp to +1 415 523 8886
    • Install zrok and create a reserved share: zrok reserve public http://localhost:8080
    • Set webhook URL in Twilio Console to: https://YOUR-TOKEN.share.zrok.io/whatsapp
  • Usage: --whatsapp-recipients +1234567890

Slack

  • Create App: https://api.slack.com/apps
  • Setup: Set SLACK_BOT_TOKEN and optionally SLACK_APP_TOKEN
  • Features: Send output to channels, receive commands
  • Usage: --slack-channel '#channel-name'

Environment Configuration

Talkito supports two environment files:

  • .env - Primary configuration (takes precedence)
  • .talkito.env - Secondary configuration (won't override .env)

Copy .env.example to .env and add your API keys:

cp .env.example .env
# Edit .env with your API keys

For WhatsApp setup with zrok tunneling:

  • ZROK_RESERVED_TOKEN: Your zrok reserved share token for webhook tunneling

Requirements

  • Python 3.10+
  • macOS (with say command) or Linux (with espeak, festival, or flite)
  • Optional: SpeechRecognition and pyaudio for ASR support
  • Optional: Provider-specific Python packages (installed as needed)

Contributing

See CONTRIBUTING.md for details on our code of conduct and the process for submitting pull requests.

License

This project is licensed under the GNU Affero General Public License v3.0 or later - see the LICENSE file for details.

Copyright (C) 2025 Robert Macrae

Metadata

Release files for talkito 1.0.1

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for talkito 1.0.1
File Size Uploaded Provenance Actions
talkito-1.0.1.tar.gz 250.7 kB Without provenance Details for talkito-1.0.1.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for talkito 1.0.1
File Interpreter ABI Platform Actions
talkito-1.0.1-py3-none-any.whl Python 3 none any Details for talkito-1.0.1-py3-none-any.whl

Total release size: 505.3 kB

Release files / talkito-1.0.1.tar.gz

Download URL talkito-1.0.1.tar.gz
Size 250.7 kB
Tags Source
SHA-256 checksum
How to use checksums
5e96fcc7823d57fbf8646da6ca2a88583201652f8e6ad24b2e50b5e71ae34f2e
BLAKE2b-256 checksum
How to use checksums
6cecf9d1f78ff405e3ebda79bf4eab3cbfba90ecc0bb3c170bd70ee64b994fef
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via twine/7.0.0 CPython/3.12.8

Release files / talkito-1.0.1-py3-none-any.whl

Download URL talkito-1.0.1-py3-none-any.whl
Size 254.7 kB
Tags Python 3
SHA-256 checksum
How to use checksums
c42dc682a37346ba056e77133cb7eda56f629106f5003dfea0cde7efb80667fa
BLAKE2b-256 checksum
How to use checksums
3fea87a29d8692b9555978d4a230f6c396048a9083fcb0ea32954c7a647bdcd4
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via twine/7.0.0 CPython/3.12.8

Security for talkito 1.0.1

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

No provenance found

None of the files in this release were uploaded with provenance metadata.

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Release history Release notifications | RSS feed

This release

1.0.1 This release

2 release files No provenance

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page