About vocode
Vocode is an open source Python library for building voice based conversational AI agents. It provides the streaming and turn based conversation abstractions, endpointing and emotion tracking that a voice agent needs, and wires them to interchangeable speech to text, language model and text to speech providers, so a developer can assemble an agent that talks over the phone, in a Zoom meeting, through a web page or against system audio.
It is aimed at developers who want to build and self host their own voice agents rather than buy a finished product. Transcription providers include AssemblyAI, Deepgram, Gladia, Google Cloud, Microsoft Azure, RevAI and Whisper; language models include OpenAI and Anthropic models, with support for LangChain agents; and synthesis providers include ElevenLabs, Cartesia, Rime, Microsoft Azure, Google Cloud, Play.ht, AWS Polly, Coqui and Bark. The library handles inbound and outbound phone calls and can be deployed to telephony, web and Zoom.
The library is free under the MIT license. Vocode also documented a hosted API for running agents on phone calls, but no pricing for it is published today.
Buyers should know the state of the project. The company website now redirects to the GitHub organization, the hosted dashboard no longer responds, the main repository was last updated in November 2024 and its maintainers say they are actively looking for community maintainers. The repository is not archived and remains usable, but this is a community sustained library rather than a supported commercial product.