Description
Vocode is an open-source Python library, published as the vocode-core repository, for building voice-based applications powered by large language models. It provides the abstractions, orchestration and integrations needed to run a real-time, streaming conversation between a person and an LLM-backed agent.
Read More
The library supports several entry points for a voice conversation: running one over a machine's system audio and microphone, standing up a phone number that answers inbound calls with an LLM agent, placing outbound calls from a managed number, or dialing into a Zoom call. It also documents an integration for using an outbound phone call as part of a LangChain agent's toolset.
Vocode is built around swappable providers rather than a single fixed stack. For transcription (speech to text), it integrates with services including AssemblyAI, Deepgram, Gladia, Google Cloud, Microsoft Azure, RevAI, and OpenAI's Whisper (both hosted and the local whisper.cpp). For the conversational agent itself, it integrates with OpenAI and Anthropic models. For speech synthesis (text to speech), it supports providers including Rime.ai, Microsoft Azure, Google Cloud, Play.ht, ElevenLabs, Cartesia, the open-source Coqui TTS, gTTS, StreamElements and AWS Polly. A companion React SDK is published separately for building web front ends around Vocode-based agents.
The project is MIT licensed, has around 3,800 GitHub stars, and is not archived. Vocode is community-oriented: its README states the project is actively looking for community maintainers and welcomes contributions of new features, integrations and documentation, coordinated through a public Discord server and a documented contribution guide.
Installation is a single `pip install vocode`, and the project's own quickstart demonstrates wiring together a Deepgram transcriber, an OpenAI ChatGPT-based agent, and an Azure speech synthesizer into a working streaming conversation in a few dozen lines of Python. Full documentation, including self-hosted telephony setup, is published at docs.vocode.dev.
- CategoryAI Voice Agents
- Pricing ModelOpen source
- DeploymentCloud
- PlatformsWeb
Vocode's Core Features
Open-source Python library for building real-time, streaming voice conversations with an LLM
System-audio mode for running a conversation through a computer's microphone and speakers
Inbound and outbound phone call support via a managed phone number
Zoom call dial-in support
LangChain integration for outbound calls from a LangChain agent
Swappable transcription providers (AssemblyAI, Deepgram, Gladia, Google Cloud, Azure, RevAI, Whisper, whisper.cpp)
Swappable LLM agent backends (OpenAI, Anthropic)
Swappable speech-synthesis providers (ElevenLabs, Cartesia, Azure, Google Cloud, Play.ht, Coqui, gTTS, StreamElements, AWS Polly)
Companion React SDK for building web interfaces around Vocode agents
MIT licensed, open to community contribution and maintainership
Getting Started with Vocode
Install the library: run `pip install vocode` in a Python environment
Configure provider keys: set API keys for the transcription, LLM and synthesis providers you plan to use
Assemble a conversation: instantiate a transcriber, an agent and a synthesizer, then wire them into a StreamingConversation object
Run locally: test the conversation over your system microphone and speakers using the provided quickstart
Add telephony: follow the self-hosted telephony guide to connect inbound or outbound phone numbers
Deploy and extend: swap in different providers, or extend the library for a custom use case such as a Zoom dial-in agent
Vocode's Use Cases
- Custom phone voice agents
- Voice assistant prototypes
- Zoom-based voice agents
- LangChain voice tools
