Vexa

Open-source, self-hosted meeting bot & transcription API.

A bot joins your Google Meet, Microsoft Teams, and Zoom calls and streams speaker-attributed transcripts in real time through an API you host — then feeds sandboxed agents that build a Markdown knowledge base your team owns. Self-hosted, Apache-2.0, air-gap-ready.

vexa.ai still runs the 0.10.6.13 line — it will host 0.12.

Why Vexa

Every meeting-AI tool you can buy sends your conversations to their cloud and rents you access back. Vexa inverts that: run the stack yourself, point it at your own models, own what your meetings become.

No one else has all three:

1. Vexa is in the meeting. A real bot joins Meet, Teams, and Zoom and streams speaker-attributed transcripts live. That bot fleet is the genuinely hard part — every "chat with your docs" tool starts after a transcript exists. Vexa produces it.

2. Your knowledge is files you own. Meetings compile into Markdown in a git repo — portable, diffable, greppable. Knowledge as code.

3. Agents work it, safely. Sandboxed coding agents read and write that repo like developers — isolated ephemeral containers, no egress, thousands in parallel, on Docker or your Kubernetes.

Only here for the transcription API? It's a complete standalone product — send a bot, read the stream, ignore the agent lane entirely.

Related MCP server: Google Meet MCP Server

Table of contents

• Quickstart

• How it works

• The agentic runtime

• Agents & your workspace

• The Terminal: AI-augmented meetings

• How-to recipes

• Deployment options

• Deploy & configure

• How Vexa is different

• For regulated enterprises

• API reference

• Status & roadmap

• Community & contributing

• License

⚡ Quickstart

Self-host the whole stack on one host, then explore it in the Terminal or drive it over the API. Linux (Ubuntu 24.04) is the production target; a Mac with Docker Desktop works fine for a local evaluation — everything runs in containers either way.

Prerequisites — make, Docker engine ≥ v26 (make all checks), and transcription: a free token at vexa.ai/account, or self-host the (GPU) transcription unit for a fully air-gapped setup. Without transcription, bots still join and record — they just produce no text (make all warns when the credentials block in .env is empty).

Build machine: The full stack (make all) requires at least 8 vCPUs and 16 GB RAM. A smaller box can run make lite (the single-container all-in-one image) but make all (Docker Compose) will likely fail or timeout. make lite is the lighter path for resource-constrained hosts.

git clone https://github.com/Vexa-ai/vexa.git && cd vexa
make all      # full Docker Compose stack — seeds .env, builds, prints your API key + URLs
make bot      # build the meeting bot from source (required before a bot can join)

When make all finishes it prints your key and URLs:

  Terminal UI : http://localhost:13000     # the web workbench
  API gateway : http://localhost:18056     # the API
  API key     : vxa_…

Explore in the Terminal (the fast path)

The Terminal is the way to see what Vexa can do. Open http://localhost:13000 — you're already signed in to a self-host account. From the workbench you can, with no curl:

• Send a bot — paste a Meet / Zoom / Teams / Jitsi URL; a bot joins as a participant.

• Watch the transcript stream in live, speaker-attributed, draft-then-confirmed.

• Chat with your workspace — ask an agent that has every captured meeting as context, and watch it commit what you decide.

Or drive it over the API

export API_KEY=vxa_...
export API_BASE=http://localhost:18056

# WIN 1 — send a bot into a live call, then read the transcript as it streams
curl -X POST "$API_BASE/bots" \
  -H "X-API-Key: $API_KEY" -H "Content-Type: application/json" \
  -d '{"platform":"google_meet","native_meeting_id":"abc-defg-hij","bot_name":"Vexa"}'

curl -H "X-API-Key: $API_KEY" "$API_BASE/transcripts/google_meet/abc-defg-hij"

# WIN 2 — ask an agent that has your whole workspace as context (answer streams back as SSE)
curl -N -X POST "$API_BASE/agent/chat" \
  -H "X-API-Key: $API_KEY" -H "Content-Type: application/json" \
  -d '{"prompt":"What did we decide in my last meeting?"}'

platform is google_meet · teams · zoom · jitsi; native_meeting_id is the code from the join URL. The agent reply streams as Server-Sent Events — message-delta frames carry the text, commit frames mark anything it recorded into your workspace.

🧩 How it works

One gateway, two domains — Meetings (capture) and Agents (work the knowledge) — both running on the same runtime: the engine that spawns every bot and every agent in its own sandboxed container.

A bot and an agent are the same runtime.v1 workload — isolated, ephemeral, reaped on idle — so the machinery already proven by thousands of meeting bots is exactly what runs your agents. Every arrow stays inside your network.

⚙️ The agentic runtime

A CLI coding agent is just a process on Linux. The runtime makes that a multi-tenant, sandboxed execution layer safe to point at real business data — the same engine that already spawns Vexa's meeting bots in production.

• Isolated. Every dispatch gets its own container: no egress except brokered tools, and only its granted workspaces exist in its filesystem — enforced by the substrate, not by the agent. Agents never run in the control plane.

• Ephemeral. A container lives while it works and is reaped on idle; continuity is a session file in the workspace. Sub-second starts, thousands in parallel.

• Orchestration-agnostic. One runtime.v1 lifecycle, pluggable substrate — the same dispatch runs identically across:

Same control plane, same worker — only how the container is created changes. One laptop to a Kubernetes/OpenShift cluster, inside your walls.

🧠 Agents & your workspace

Capture is the front door; agents make the knowledge compound. Every meeting compiles into your workspace — a git repo of Markdown (an Open Knowledge Format kg/ bundle) that agents (Claude Code, Codex, …) read and write like developers work a codebase.

This is Andrej Karpathy's LLM Wiki pattern, run as a team service. The idea: don't RAG over raw documents — where the model rediscovers everything from scratch on every question — have agents compile sources into structured, interlinked markdown entity pages (people, companies, projects, decisions) so knowledge compounds. Vexa builds that wiki for you from the richest source there is: your meetings. Each call is ingested into entity pages; agents keep them current between calls; every answer starts from what your team already knows — on your own servers.

Agents work any workspace; a meeting is just one trigger of four — chat, schedule (cron), event (e.g. incoming email), finished meeting. Meetings themselves are scheduled work: connect your calendar (ICS) and planned meetings appear with attendees — bots auto-join, agents prepare before the call and process after it.

• Multiplayer. Team-shared, attributed workspaces — not one person's private notes.

• Automated. The bot captures the call; the transcript compiles itself in.

• Safe by design. Agents are untrusted and enforce nothing themselves. You, in chat, write directly (git is the undo); untrusted input — an email, a web page — runs propose-only: the agent suggests, a human approves, trusted code applies. Irreversible effects are always gated.

Status (honest): capture, transcription, and speaker attribution are production; the agent dispatch core is built and proven live end-to-end. What's still landing is tracked in Status.

🖥️ The Terminal: AI-augmented meetings

0.12 ships a new Terminal UI built to put the backend's scale — thousands of bots and agents — to work on your actual week. It opens on your meetings: coming up, live now, to review.

• An agent in your meeting, with your knowledge. Open a live call: the transcript streams speaker-attributed, and the agent has the live conversation and your workspace in context. Ask mid-call "what did we promise them last time?" — or research a person, company, or contract the moment it comes up, grounded in your wiki.

• Knowledge built on meetings — and between them. Every planned meeting gets an agent that prepares the brief before (who's coming, history, open threads — it interviews you for what it can't know) and processes the transcript after. Arrive prepared, leave with the wiki updated.

• Sharing. Invite colleagues into a workspace — same wiki, attributed. Share a meeting with its attendees — they get the real-time feed, not a recording link after the fact.

• Collaborative, AI-augmented meetings. Prep a shared workspace together; during the call, humans edit the brief while agents stream the transcript in and work the knowledge — one room, human and AI participants on the same files.

📖 How-to recipes

Each is a complete path to one outcome over the Agent API. Full guides at docs.vexa.ai.

💬 Chat with your workspace — ask an agent that has every meeting, email, and note as context; trusted chat can also record a decision (a git commit).

curl -N -X POST "$API_BASE/agent/chat" -H "X-API-Key: $API_KEY" -H "Content-Type: application/json" \
  -d '{"prompt":"Brief me on the Acme account: every meeting, the open decisions, and the next step."}'

🌅 Brief me every morning — an unattended agent on a cron schedule that commits to your workspace.

curl -X POST "$API_BASE/agent/routines" -H "X-API-Key: $API_KEY" -H "Content-Type: application/json" \
  -d '{"name":"Morning brief","cron":"0 8 * * 1-5",
       "prompt":"Brief me from overnight activity — new meetings, decisions, follow-ups due. Write brief/today.md.",
       "run_now":true}'

📝 Report after every meeting — dispatch a one-shot agent when a call ends (or a routine that sweeps recent meetings).

curl -X POST "$API_BASE/agent/invocations" -H "X-API-Key: $API_KEY" -H "Content-Type: application/json" \
  -d '{"runner":"claude-code","workspaces":[{"id":"u_jane","mode":"rw"}],"trigger":"scheduled",
       "start":{"entrypoint":{"inline":"Write a report for the meeting that just ended: summary, decisions, action items with owners."}}}'

📧 Triage incoming email (safely) — an event-triggered agent that gets the mailbox read-only and can only propose actions as cards; a human approves before anything is written or sent.

curl -X POST "$API_BASE/agent/events" -H "X-API-Key: $API_KEY" -H "Content-Type: application/json" \
  -d '{"name":"email.received","source":{"uri":"mailbox://u_jane/INBOX/AB12CD"},
       "plan":{"prompt":"Triage this email into tasks; propose a record for each action item and a draft reply."}}'

Live-meeting copilot — cards for people, decisions, and action items during the call (POST /agent/meeting/start → stream GET /agent/meeting/stream) — is on the roadmap; see Status.

🚀 Deployment options

Two ways to run Vexa, one codebase:

1. Personal / dev — Docker on your Mac, Linux, or Windows machine. Single container (make lite — the all-in-one Vexa Lite image) or the full Compose stack (make all). Reuse your Claude subscription: workers run the official claude CLI against your own Pro/Max credential, which is a covered, credit-metered use under Anthropic's terms for a personal deployment — your subscription, your turns, your machine. See Model credentials & licensing for the exact terms mapping (Anthropic's Agent SDK plan-usage article is the primary source). You get the full service — bots, transcripts, agents, Terminal — on the subscription you already pay for.

2. Cloud — Helm on Kubernetes / OpenShift, scalable to thousands of users. The chart in deploy/helm deploys the same control plane with RUNTIME_BACKEND=k8s: every bot and every agent is its own Kubernetes workload (a bare Pod per dispatch), so capacity is your cluster's scheduler, not a bigger box — built multi-tenant and multiuser from the start. One compliance rule when you go multiuser: other users' turns must run on an API key (Commercial Terms), never a personal subscription credential — the licensing page spells out the boundary, and Settings → Models enforces per-user/global credential resolution. K8s backend status is tracked honestly in Status.

🏠 Deploy & configure

make all brings up the full stack via Docker Compose on one Linux host — each service in its own container, bound to loopback:

• Runtime backend — RUNTIME_BACKEND=docker (default) or k8s (a Pod per dispatch).

• Transcription is a separate GPU unit — make all runs GPU-free; stand up the STT service (faster-whisper, OpenAI-compatible) from deploy/transcription on any GPU box and point .env at it. Or use a free hosted token at vexa.ai/account while testing.

• Bring your own inference — point the agent at your own LLM endpoint; no inference leaves the network.

• Air-gapped — everything in-VPC, zero egress — the posture the regulated verticals require.

• Targets — make all · make bot (build the bot image from source — required, not pulled) · make lite · make up / make down · make help. Expose the Terminal via a TLS reverse proxy for production; full guide in the docs.

🆚 How Vexa is different

The crowded "AI second brain / self-hosted knowledge base" space is full of excellent tools for reasoning over documents you already have. None of them join a live meeting — they consume transcripts other tools produced. That's the whole point: capture is the moat, and it sits upstream of where a document-RAG tool's architecture even starts.

Against the tools developers actually weigh for meeting capture:

Vexa is the one combination the others don't offer: a permissively-licensed (Apache-2.0) meeting-bot-API server that is self-hosted × real-time × multi-platform × knowledge-you-own. And it's complementary to the document-RAG and "second brain" tools — feed them Vexa's clean, attributed transcripts and let them do what they're good at.

The full field — including Attendee (the other open-source meeting-bot API) and the local-notetaker tools — is mapped honestly, trade-offs and all, in How Vexa compares.

🏦 For regulated enterprises

For banks, healthcare, government, and anyone in a regulated industry, the meeting-AI question isn't "which cloud" — it's "how do we get this without a cloud." Vexa is air-gapped meeting intelligence — the sovereign alternative to Microsoft Copilot — built for exactly that buyer.

You don't compete with a notes app here — you replace Microsoft 365 Copilot and Zoom AI Companion on the axes they structurally can't move:

What that means in practice:

• Air-gapped — fully offline, your infrastructure, your models. Nothing phones home.

• Adaptive — your engineers implement requirements directly: domain vocabulary, underserved languages, custom workflows. No vendor feature queue.

• Owned, not rented — deploy once, extend without asking permission. No per-seat tax.

• Scales inside your walls — thousands of isolated agents in parallel on Docker or your Kubernetes/OpenShift cluster.

Evaluate it for your org — the artifacts a security review asks for, in this repo:

Full review page: Security & compliance in the docs.

Regulated banks and Fortune-500s run Vexa fully air-gapped on their own OpenShift and local LLMs today.

📡 API reference

Two APIs behind the gateway, authenticated with X-API-Key. Base URL: http://localhost:18056 (self-host) or https://api.cloud.vexa.ai (hosted).

Meetings API — capture; usable standalone:

Agent API — the control plane, under the /agent/* prefix (identity is derived from your key, server-side):

platform ∈ google_meet · teams · zoom · jitsi. Full reference: docs.vexa.ai.

v0.12 note: live bot-control — PUT /bots/{…}/config (change language/task mid-call) and POST /bots/{…}/speak (TTS into the call) — plus the live-meeting copilot (/agent/meeting/*) and WebSocket streaming are not yet wired in the open-core stack and return 404 today. Send-a-bot, stop, status, transcripts, recordings, agent chat, routines, and events are live.

🗺️ Status & roadmap

Honest state of the 0.12 line (mirrors the status page — never aspirational):

✅ Production · 🟡 In progress · 🔵 Planned

🤝 Community & contributing

• Docs — docs.vexa.ai

• Discord — discord.gg/Ga9duGkVz9

• Roadmap — the board, grouped by contributor lane, with milestones as the version gates.

• Contributing — how delivery works: prepared issues with acceptance tables that guarantee merge, and human validation credited as a first-class contribution (one page, law and how-to together).

• Issues & PRs — welcome. See SECURITY.md to report vulnerabilities.

Vexa is built in the open. If you self-host it, extend it, or run it air-gapped somewhere interesting, we'd love to hear about it.

📄 License

Apache-2.0. Own it, run it, fork it, ship it. It's an investment, not a rental.