Vexa

Vexa is an open-source, self-hostable meeting bot API and meeting transcription API for Google Meet, Microsoft Teams, and Zoom. Alternative to Recall.ai, Otter.ai, and Fireflies.ai — self-host so meeting data never leaves your infrastructure, or use vexa.ai hosted.

Data sovereignty — self-host so meeting data never leaves your infrastructure

Cost — replace $20/seat SaaS with your own infrastructure

Embed in your product — multi-tenant meeting bot API with scoped tokens

AI agents — MCP server with 17 tools

Capabilities

Every feature is a separate service. Pick what you need, skip what you don't. Self-host everything or use vexa.ai hosted.

Why Self-Host Meeting Transcription?

For regulated industries — banks, financial services, healthcare — meeting data can't leave your infrastructure. Self-hosting Vexa means zero external data transmission and full audit trail on your own infrastructure.

For cost-conscious teams — replace per-seat SaaS pricing. A team paying $17/seat/mo for meeting transcription can self-host Vexa and drop that to infrastructure cost.

For developers — embed a meeting bot API in your product. Multi-tenant, scoped API tokens, no per-user infrastructure.

Build meeting assistants like Otter.ai, Fireflies.ai, or Fathom — or build a meeting bot API like Recall.ai — self-hosted on your infrastructure.

• Vexa (self-host) — your infra cost. Data never leaves your infrastructure. Meeting bot API, real-time transcription, interactive bots, MCP server. Open source, Apache 2.0. Google Meet, Teams, Zoom.

• Recall.ai — $0.50/hr. No self-hosting. Meeting bot API, real-time transcription. No MCP, limited interactive bots. Closed source. Meet, Teams, Zoom, Webex.

• Otter.ai — $17-20/seat/mo. No self-hosting. No API. Limited real-time transcription. Closed source. Meet, Teams, Zoom.

Or use vexa.ai hosted — get an API key and start sending bots immediately, no infrastructure needed.

Built for Data Sovereignty

Meeting data never leaves your infrastructure. Self-host for complete control. Modular architecture scales from edge devices to millions of users.

1. Hosted service At vexa.ai — get an API key and start sending bots. No infrastructure needed. Ready to integrate

2. Self-host with Vexa transcription Run Vexa yourself, use vexa.ai for transcription — ready to go, no GPU needed. Control with minimal DevOps — see deploy/ for setup guides.

3. Fully self-host Run everything including your own GPU transcription service. Meeting data never leaves your infrastructure — see deploy/ for setup guides.

What's new

v0.10.4

• Zoom (via web) — platform=zoom works out of the box, no SDK setup.

v0.10 — full architecture refactor

• Services refactored — runtime-api as infrastructure layer (container orchestration), meeting-api as data layer, agent-api as high-level intelligence layer. Clean separation of concerns.

• Real-time pipeline moved into bots — transcription pipeline now runs inside bot containers, eliminating external dependencies

• Agent API (experimental) — ephemeral containers for AI agents. See services/agent-api/.

• Helm/K8s — production Kubernetes deployment with built images and global.imageTag support

v0.9

• Zoom (experimental) — initial Zoom Meeting SDK support

• Interactive Bots API — speak, chat, screen share, avatar controls during live meetings

• MCP server — 17 tools for AI agents

• Recordings — S3-compatible storage

See full release notes: https://github.com/Vexa-ai/vexa/releases

Quickstart

Self-host with Docker

On a fresh Linux machine (Ubuntu 24.04):

apt-get update && apt-get install -y make git curl
curl -fsSL https://get.docker.com | sh
git clone https://github.com/Vexa-ai/vexa.git && cd vexa

Then choose:

Both prompt for a transcription token on first run. Get one at vexa.ai/account, or self-host transcription with a GPU.

Guides: Vexa Lite | Docker Compose | Helm (K8s)

Hosted (no deployment needed)

Get your API key at vexa.ai/account and start sending bots immediately.

Meeting API — Send Bots, Get Transcripts

Send a bot, get real-time transcripts with per-speaker audio and interactive controls (speak, chat, share screen).

# Send a bot to Google Meet
curl -X POST "$API_BASE/bots" \
  -H "Content-Type: application/json" \
  -H "X-API-Key: <API_KEY>" \
  -d '{"platform": "google_meet", "native_meeting_id": "abc-defg-hij"}'

# Get transcripts
curl -H "X-API-Key: <API_KEY>" \
  "$API_BASE/transcripts/google_meet/abc-defg-hij"

Works with Google Meet, Microsoft Teams, and Zoom. Set API_BASE to https://api.cloud.vexa.ai (hosted) or http://localhost:8056 (self-hosted).

For real-time WebSocket streaming, see the WebSocket guide. For full REST details, see the User API Guide.

Browser Bots — Persistent Browser Containers for Agents

Remote browser containers with CDP + Playwright access and persistent session storage via S3. Agents get a real browser that stays logged in across restarts — Google, Microsoft, or any web session.

• CDP + Playwright — full browser automation via Chrome DevTools Protocol

• Persistent sessions — authenticated browser state saved to S3, restored on next spin-up

• VNC access — humans can observe and control the browser in real time alongside agents

• On-demand containers — spin up in seconds, auto-reclaim when idle

See features/browser-session/ and features/remote-browser/ for details.

MCP Server — Meeting Tools for AI Agents

17 tools that let AI agents join meetings, read transcripts, speak, chat, and share screen. Works with Claude, Cursor, Windsurf, and any MCP-compatible client.

Your AI agent can join a meeting, listen to the conversation, and participate — all through MCP tool calls. See services/mcp/ for setup and tool reference.

Modular — Pick What You Need

Vexa is a toolkit, not a monolith. Every feature works independently. Use one or all — they compose when you need them to.

You don't pay complexity tax for features you don't use. Each service is a separate container. Don't need agents? Don't run agent-api. Don't need TTS? Don't run tts-service. Services communicate via REST and Redis, not tight coupling.

Roadmap

For the up-to-date roadmap and priorities, see GitHub Issues and Milestones. Issues are grouped by milestones to show what's coming next, in what order, and what's currently highest priority.

• Issues: https://github.com/Vexa-ai/vexa/issues

• Milestones: https://github.com/Vexa-ai/vexa/milestones

For discussion/support, join our Discord.

Architecture & Feature Status

Each service and feature has its own README with architecture, DoD table, and evidence-based confidence scores.

• Services: api-gateway • meeting-api • admin-api • runtime-api • vexa-bot • transcription-service • tts-service • mcp • dashboard • agent-api (experimental)

• Features: realtime-transcription • bot-lifecycle • browser-session • remote-browser • speaking-bot • meeting-chat • webhooks • authenticated-meetings

• Deploy: Docker Compose • Vexa Lite • Helm/K8s

• Guides: Vexa Lite Deployment • Docker Compose Deployment • Self-Hosted Management • Recording Storage

Contributing

We use GitHub Issues as our main feedback channel — triaged within 72 hours. Look for good-first-issue to get started. Join Discord to discuss ideas and get assigned.

Links

Website • Docs • Discord • LinkedIn • X (@grankin_d) • Meet Founder

Related: vexa-lite-deploy • Vexa Dashboard