Remote Agents with OpenVoiceOS

While OpenVoiceOS is designed primarily for local-first usage, more advanced deployments—like hosting agents in the cloud, connecting multiple voice satellites, or enabling multi-user access through a web frontend—are made possible via the HiveMind companion project.

HiveMind Server

HiveMind is a distributed voice assistant framework that allows you to expose AI Agents (either full ovos-core installs or just individual personas) over a secure protocol.

💡 Unlike the lightweight persona-server, HiveMind is designed for trusted, networked setups.

Key Features:

  • Secure Access: Communicates over the HiveMind protocol, which supports authentication, encryption and granular permissions — safe for exposing OVOS to remote clients or satellites.
  • Agent Plugins: Agent plugins integrate the HiveMind protocol with various frameworks, including OpenVoiceOS. Keep your existing infrastructure even when you totally replaces the brains!
  • Multi-User Ready: Great for use in cloud hosting, web portals, or enterprise environments where access control is critical.
  • Composable: Let local personas delegate questions to a smarter remote OVOS instance.

Typical Use-cases:

  • 🏡 Running OpenVoiceOS on a powerful server or in the cloud.
  • 🛰️ Connecting lightweight devices (satellites).
  • 📱 Remote access to OpenVoiceOS.
  • 🧑‍🤝‍🧑 Serving multiple users or applications concurrently.

Check out the HiveMind documentation for more info

HiveMind Personas

The hivemind-persona-agent-plugin project allows you to expose a single persona—not the full OVOS stack—through hivemind

This enables you to deploy AI agents for external use without needing a full OVOS assistant.

Untitled-2025-04-15-2340(15)

Why Use It?

  • Minimal attack surface (persona only, no full assistant features).
  • Can be queried remotely using the HiveMind protocol.

💡 This is not the same as persona-server. hivemind-persona-agent-plugin uses a secure protocol (HiveMind), while ovos-persona-server uses insecure HTTP.

Server Configuration

in your hivemind config file ~/.config/hivemind-core/server.json

{
  "agent_protocol": {
    "module": "hivemind-persona-agent-plugin",
    "hivemind-persona-agent-plugin": {
      "persona": {
      "name": "Llama",
      "solvers": [
        "ovos-solver-openai-plugin"
      ],
      "ovos-solver-openai-plugin": {
        "api_url": "https://llama.smartgic.io/v1",
        "key": "sk-xxxx",
        "persona": "helpful, creative, clever, and very friendly."
      }
    }
    }
  }
}

HiveMind as a Solver Plugin

Want your local assistant to ask a remote one when it's stuck? You can!

The hivemind-bus-client can function as a solver plugin, allowing you to:

  • 🦾 Delegate processing to a more powerful/secure server for specific tasks.
  • 🌍 Handle outages: Handle intermitent local agent failures from other solver plugins in your persona definition
  • 🤝 Use remote hivemind agents in a collaborative AI / MoS (mixture-of-solvers) setup.

Untitled-2025-04-15-2340(14)

🤖 “When in doubt, ask a smarter OVOS.”

For usage with persona, use "ovos-solver-hivemind-plugin" for the solver id

{
  "name": "HiveMind Agent",
  "solvers": [
    "ovos-solver-hivemind-plugin"
  ],
  "ovos-solver-hivemind-plugin": {"autoconnect": true}
}

You can also use it in your own python projects

from ovos_hivemind_solver import HiveMindSolver

bot = HiveMindSolver()
bot.connect()  # connection info from identity file
print(bot.spoken_answer("what is the speed of light?"))

Chaining Components for Flexible Deployments

HiveMind and persona-server can be combined to bridge secure and insecure environments, depending on your needs:

  • expose existing OpenAI/Ollama servers to hivemind satellites securely
    • connect hivemind satellites directly to existing LLM apps (eg. ollama)
  • expose a remote hivemind-core to local insecure ollama/openai endpoints
    • eg. to integrate hivemind into HomeAssistant
  • expose a localhost ovos-core/persona.json to local insecure ollama/openai endpoints
    • half-way compromise, does not expose the full messagebus and does not require hivemind
    • easier to setup and configure
Use Case Tool Secure? API Type Notes
Local interface + Persona ovos-persona-server + persona.json ❌ OpenAI-compatible Great for quick setups, not public exposure`,HTTP, no auth
Local interface + OpenVoiceOS ovos-persona-server + ovos-solver-bus-plugin ❌ OpenAI-compatible OpenVoiceOS bus must be exposed to ovos-persona-server,HTTP, no auth
Local interface + HiveMind Agent ovos-persona-server + ovos-solver-hivemind-plugin ❌ OpenAI-compatible Same as above, but for any remote hivemind agent,HTTP, no auth
Secure remote OpenVoiceOS agent hivemind-core + hivemind-ovos-agent-plugin + ovos-core ✅ HiveMind protocol Auth, encryption, granular permissions, HTTP or Websockets
Secure remote Persona agent hivemind-core + hivemind-persona-agent-plugin + persona.json ✅ HiveMind protocol Auth, encryption, granular permissions, HTTP or Websockets

The first 3 examples allow us to integrate our Agents with HomeAssistant via the Ollama Integration

The last 2 examples allow us to integrate with HiveMind ecosystem and all the existing satellite implementations

While useful for experimentation, some other persona access methods are not secure for remote use:

🦙 ovos-persona-server:

  • ✅ Compatible with OpenAI/Ollama APIs.
  • ❌ HTTP only, not encrypted or authenticated.
  • ✅ Useful to expose personas to HomeAssistant, OpenWebUI, and similar local network tools.

🏠 HomeAssistant + ovos-persona-server:

  • 🗣️ Can route HomeAssistant wyoming satellites to an OVOS persona.
  • ❌ Uses Wyoming protocol, which lacks hivemind's security features.