Vexa Main Project Repo
The first real‑time meeting co‑pilot that listens, thinks, and speaks to actively assist your Google Meet / Zoom / MS Teams calls.
Most "AI note‑takers" just sit in a call and dump a doc later. Hey Vexa is the first open‑source bot that actively joins the conversation. Say "Hey Vexa, research this problem and come back with the report" (or any custom command) and the agent will
1. Invoke an agent with access to your favorite MVP tools (previous meetings, RAG, company docs, web search, etc)
2. Response with a natural voice that answers back into the meeting — instantly.
Built on top of the Vexa real‑time transcription API, which only requires the addition of an extra endpoint. It adds just one missing piece: a lightweight playback endpoint so the bot can talk.
effort: light
- update
@vexa-botto playback sound, obtained through Redis MQ so an external service can streamaudio/wavback into the call. - Extend @bot-manager with
/playbackendpoint to queue the audio playback jobs via Redis
effort: easy
- Transcript Listener – poll Vexa's real‑time caption stream.
- Wake‑word Detector – regex for "Hey, Vexa" triggers an agent.
- Simple LLM one pass response
- TTS speech generation
- Agent Invocation – send recent context → LLM → TTS →
POST /playback.
effort: should be easy
- Replace LLM call with the MCP-enabled agent
We're looking for backend hackers, prompt engineers, TTS/LLM enthusiasts, DevRel storytellers, and testers to help build this in the open. 👉 Hop into the Discord, pick an issue, and say "Hey Vexa, I'm in!"