Build streaming voice LLM agents with Vocode
Use Vocode to compose transcription, LLM, speech synthesis, and telephony components into reviewable real-time voice-agent workflows.
npx skills add agentskillexchange/skills --skill build-streaming-voice-llm-agents-with-vocode
Vocode is a strong fit when the operator needs to assemble a voice-agent workflow from documented Python components. The agent can configure streaming microphone input, transcription providers, a ChatGPT-style conversation agent, speech synthesis, and deployment targets such as system audio, phone calls, Zoom meetings, or LangChain-triggered outbound calls.
What this skill actually does
Invoke this instead of using a hosted voice-agent product when the user needs source-visible provider choices, custom conversation behavior, telephony routing, or local experimentation with voice interactions. The scope boundary is voice LLM orchestration: connect audio input and output, select ASR/TTS/LLM integrations, run a conversation loop, and inspect call or transcript behavior. It should not be treated as a generic Python package listing or a broad voice platform card. Operators should keep credentials, phone-number operations, recordings, and user consent requirements explicit before live calls.
Inputs and prerequisites: Python environment with the vocode package; provider credentials for selected transcription, LLM, speech synthesis, and telephony services; optional LangChain integration for agent-triggered outbound calls..
Setup notes: Install the upstream Python package with pip install vocode, then follow the open-source quickstart to configure microphone or telephony input, a conversation agent, transcriber, and synthesizer.
Source and verification boundary: use https://docs.vocode.dev/open-source as the canonical reference before running the workflow; keep commands, API calls, CLI usage, and generated outputs reviewable against that upstream source.
Framework fit: publish this as a Custom Agents workflow only when the operator can invoke the documented toolchain directly, rather than treating the upstream project as a generic product listing.