hivemind-core Server¶
Everything in a HiveMind points back to one machine: the server in the closet, on the desk, in the container. hivemind-core is what accepts every satellite's connection, decides which AI actually answers, and holds the line on who's allowed to do what. The satellites are interchangeable; this is the piece that makes them a hive. This section is for whoever runs it.
Which server do I want?
A server's "personality" is set by one choice: which agent plugin answers requests. Pick by what you want the assistant to be:
- A full voice assistant with skills → OVOS Skills Server
- A chatbot / LLM (no skills needed) → Persona Server
- An external AI agent (A2A protocol) → A2A Server
Not sure, or want the easy path? The Admin Panel is a web UI that
launches a hivemind-core instance and lets you administer it from the browser — one command.
What's in this section¶
| Page | What it covers |
|---|---|
| Admin Panel | A web UI that launches a hivemind-core instance and administers it — the easiest way to run one. |
| OVOS Skills Server | The default setup: serves OpenVoiceOS skills, STT, and TTS. |
| Persona Server (LLM) | Serve an LLM/chatbot persona — no ovos-core required. |
| A2A Server (Agents) | Bridge to any agent speaking Google's Agent-to-Agent protocol. |
| Transports | How bytes travel: WebSocket (default), HTTP, MQTT, Usenet. |
| Audio Binary Protocol | How raw audio (wake word, STT, TTS streams) travels to/from thin satellites. |
| Docker Deployment | Run hivemind-core and the OVOS stack in containers. |
| Operations | Day-2 concerns: TLS, reconnection, monitoring, connection errors. |
How a request flows through hivemind-core¶
Whichever flavour you pick, every question a satellite asks takes the same short journey through the server. Follow one utterance left to right:
satellite ──▶ transport plugin ──▶ permission check ──▶ agent plugin ──▶ AI backend
(mic, CLI) (WebSocket/HTTP…) (allowed? ) (OVOS/LLM/A2A) (skills/LLM)
Three checkpoints, three questions: the transport decides how the bytes arrive, the permission check decides whether the message is allowed through, and the agent plugin decides who answers it. What makes hivemind-core hivemind-core is that all three are swappable parts, not fixed wiring — which is the whole story of Plugin Architecture.
Next: OVOS Skills Server · or Quick Start to stand one up now.