Vexa – Open‑source meeting bots & real‑time transcription

What it does (in plain language)

• Bot joins video calls – Vexa can act as a participant in Google Meet, Microsoft Teams, Zoom and (soon) Jitsi. While in the call it captures what each speaker says and produces a speaker‑attributed transcript in real time.
• Transcripts become a knowledge base – Every meeting is stored as Markdown files in a Git repository you own. The repo is version‑controlled, searchable and can be inspected with normal developer tools.
• Sandboxed AI agents – Separate “coding agents” (e.g. Claude‑Code, Codex) run in isolated containers. They read the Markdown repo, make changes (e.g. summarise a meeting, create a brief, update entity pages) and commit the results back to the repo. All work happens inside your network – no data leaves your premises.
• API + web UI – A single API gateway exposes endpoints for:
  • spawning a bot (POST /bots),
  • streaming live transcripts, and
  • chatting with an agent that has the whole workspace as context. The bundled Terminal UI (http://localhost:13000) lets you drive these features from a browser.
• Deploy anywhere – The stack runs in Docker containers on a single host, or as individual Pods in Kubernetes/OpenShift. A “lite” single‑container image is also provided.
• Air‑gap ready – All components can be kept inside a VPC or isolated network; you can also self‑host the speech‑to‑text (STT) service.

Who might use it

• Companies that need full control over meeting data (regulated industries, enterprises with strict compliance).
• Teams that want a self‑hosted “second brain” built directly from their live conversations.
• Developers who want to extend the system – plug in their own LLM endpoint, add custom agents, or integrate with calendars, email, etc.

Core technical pieces (from the README)

Component	Role
Gateway (:18056)	Auth, routing, and request validation.
Terminal UI (:13000)	Web workbench for sending bots, watching transcripts, chatting with agents.
Meeting‑API	Handles bot lifecycle and stores raw/processed transcripts.
Agent‑API	Dispatches sandboxed agents, supports chat, routines, events.
Runtime	Spins up a container (Docker, plain process, or K8s pod) for each bot or agent.
Postgres / Redis	Metadata, job queue, and key storage.
Storage	Filesystem volume where the Markdown Git repo lives.

How you get started

1. Hosted quick‑start – Sign‑up at https://vexa.ai, obtain an API key and POST a bot into a call.
2. Self‑hosted – Clone the repo, run make all (Docker‑Compose) or make lite (single image). The command prints a local API key and URLs for the Terminal and API.
3. Interact – Use curl examples from the README to:
   • spawn a bot,
   • stream the live transcript,
   • send a chat prompt to an agent, or
   • schedule routine agents (e.g., a daily brief).

What’s already production‑ready vs. upcoming

• Production: real‑time bot joining, speaker‑attributed transcription, basic agent dispatch.
• In progress: live validation for Jitsi, full K8s deployment maturity, some advanced agent features.

License & community

• Apache‑2.0 (permissive, commercial‑friendly).
• Discord community for support and discussion.
• Open source code, Docker images, Helm chart, and CLI tools are all in the same repository.

---

All details above are taken directly from the project's README; no additional assumptions have been made.