vexa: Capture and Transcribe Online Meetings

Overview

Vexa is a meeting-capture and transcription service for developers and teams that need bots to join online calls and make transcripts available through an API. It supports Google Meet, Microsoft Teams, and Zoom; Jitsi is built but still awaiting live validation.

Use the hosted service for bot capture, or self-host Vexa to keep the stack and meeting data in your own environment. The self-hosted stack also includes agents that work with meeting knowledge stored as Markdown, making Vexa relevant when transcription is part of a broader internal knowledge workflow.

Key Features

• Send bots to meetings on Google Meet, Microsoft Teams, and Zoom.
• Retrieve transcripts with speaker attribution through the API.
• Stream transcript updates through Redis; the documented open-core transcript API uses polling, not WebSocket multiplexing.
• Store recordings in your own object storage.
• Run the meeting and agent workloads using Docker, a process backend, or Kubernetes.
• Connect self-hosted agents to Markdown workspaces for chat, scheduled routines, and event-triggered tasks.
• Access an MCP endpoint for agent clients.

Use Cases

• Developers can add meeting capture and transcript retrieval to an application without building and operating their own meeting bots.
• Teams that handle sensitive conversations can self-host the capture stack and keep transcripts and recordings within their infrastructure.
• Internal knowledge teams can turn meeting transcripts into Markdown workspace material and use agents to prepare briefs or summarize follow-ups.
• Platform teams can run meeting bots and agent workloads on Docker or Kubernetes when they need control over deployment and runtime isolation.

Project Facts

• Language: Python
• License: Apache-2.0
• Stars: 2.8k
• Forks: 495
• Topics: api, claude-code, codex, google-meet, mcp, mcp-server, meeting-assistant, meeting-minutes, meeting-notes, model-context-protocol, ms-teams, notetaker, self-hosted, speech-to-text, transcription, zoom
• Archived: No

Getting Started

For hosted capture, create an account and API key at vexa.ai, then follow the send-a-bot guide. To self-host the full stack:

git clone https://github.com/Vexa-ai/vexa.git && cd vexa
make all

See the README for setup requirements, configuration, and API details.

Considerations

• The self-hosted full stack requires Docker Engine 26 or later and make; the README identifies Ubuntu 24.04 as the production target. A local evaluation can run with Docker Desktop.
• Self-hosted transcription requires a hosted transcription token or a separately deployed GPU-backed transcription service. Transcription is required by default, and bot creation returns 503 if it is unavailable.
• Jitsi support is built but live validation is pending. WebSocket transcript multiplexing and mid-call bot configuration and speech are not available in the open-core stack; the README says those control endpoints return 404.
• The agent features are part of the self-hosted stack, not the hosted service. Some workspace storage capabilities are still in progress, and at-rest encryption is listed as planned.