The open-source escape pod from $20/seat meeting SaaS

What it does Vexa deploys browser-based bots into Google Meet, Microsoft Teams, and Zoom calls. They record, transcribe per-speaker audio via Whisper, and return real-time transcripts over WebSocket or REST. You can also make them speak, chat, share screen, or change avatar mid-call. Everything runs as independent containers—take the transcription API, the MCP server, or the full stack.

The interesting bit The project treats browser automation as infrastructure. Bots run inside CDP + Playwright containers with persistent S3-backed sessions, so they stay logged in across restarts. There’s also an experimental Agent API that spins up ephemeral containers for AI agents, and an MCP server with 17 tools so Claude or Cursor can literally join your standup.

Key highlights

• Three deployment modes: fully hosted, self-hosted with hosted transcription, or fully self-hosted including GPU transcription
• Multi-tenant with scoped API tokens, designed for embedding in other products
• Real-time pipeline now runs inside bot containers (v0.10 refactor), cutting external dependencies
• Browser sessions persist to S3; VNC access for human oversight
• Next.js dashboard included; Helm/K8s manifests for production

Caveats

• Zoom support is newer and marked experimental in places; web-based Zoom works without SDK setup, but Meeting SDK support arrived later
• The Agent API is explicitly labeled experimental
• Self-hosting transcription requires a GPU or a paid token from the hosted service

Verdict Worth evaluating if you’re building meeting intelligence into your product or your compliance team frowns at Otter.ai’s data residency. Skip it if you just need occasional personal transcripts—setup overhead only pays off at scale or under regulation.