Home Assistant can send speech recognition and synthesis to local services over the Wyoming Protocol, and conversation to a local Ollama server. These are configured through Home Assistant integrations. The current Habab distribution does not include these services or models; install and choose them separately on the Home Assistant server or another computer on your home network.

This is a preliminary setup path. Whisper, Piper, and Ollama versions, the Russian voice, a model for controlling devices, and hardware requirements still need to be checked together on the target server. Do not treat the setup as offline until it has been checked with WAN disconnected.

In a source review on October 26, 2026, Habab's main Compose profile contained Home Assistant only; it did not include Wyoming or Ollama services. This article describes connecting separate local equipment and is not a report of a completed voice test. It was prepared with AI assistance and checked against Home Assistant documentation.

Key facts

• The current Habab distribution does not include Wyoming services or local models.
• Speech recognition, conversation, and speech synthesis are connected as separate local Home Assistant services.
• The home network must remain available when internet access is disconnected.
• Model suitability and speed on the target server have not been confirmed.

What you need

• Home Assistant Core with access to integration and Assist pipeline settings.
• A local Wyoming speech recognition service, such as Whisper.
• A local Wyoming speech synthesis service, such as Piper, with a Russian voice model installed.
• Ollama with a local conversation model if you want a neural network to handle conversation. Controlling entities requires a model that supports the needed tool calls; the Ollama integration marks Home Assistant control as experimental.
• Home network access to all three services. LAN is required; “without internet” does not mean “without a network.”

Check model storage and memory requirements for your chosen hardware first. Download size is not the same as runtime RAM use. Habab does not yet publish a verified hardware configuration or response time for this setup.

Local endpoint template

Enter an address that is reachable from the Home Assistant container. The ports below match common values in the separate QA profile; check the actual port used by your service.

Service	Address in Home Assistant
Wyoming Whisper STT	<LAN-hostname-or-IP>:10300
Wyoming Piper TTS	<LAN-hostname-or-IP>:10200
Ollama	http://<LAN-hostname-or-IP>:11434

These are home network addresses, not public domains. If a service runs on another computer, first check that Home Assistant can reach the listed port. Container names such as whisper, piper, and ollama work only inside a Compose network that includes those services.

Connect the local services

1. Install your chosen Wyoming STT and TTS services and Ollama on the Home Assistant server or another computer on your home network. Start them and confirm they are reachable on the LAN before configuring Home Assistant.
2. In Home Assistant, open Settings → Devices & services and add the Wyoming Protocol integration for each speech service. Enter its local address and the port from that service's configuration.
3. Add the Ollama integration and connect it to your local Ollama server. Download the model in advance and select it in the integration settings. If you enable device control, start with one test entity and check which entities the agent can access.
4. Under Settings → Voice assistants, create an Assist pipeline. Select the local Wyoming STT, local conversation agent, and Wyoming TTS. Confirm that each field points to the service you prepared.
5. In the Habab panel, choose Home Assistant Assist mode and select this pipeline. Habab mode continues to use the configured Habab route; choosing an Assist pipeline does not make that route local.

Menu names can differ between Home Assistant versions. If a service is not discovered automatically, add it manually and use its address on your network. Do not copy Home Assistant OS Apps instructions into this build: Habab Home runs Home Assistant Core in Docker without Supervisor.

Check it with WAN disconnected

Disconnect internet access at the router while leaving the home network and local addresses available. Restart Home Assistant and the selected services, then speak a short command in Russian.

Check the recognized text, the selected test entity's state change, the local conversation model's reply, and Piper speech separately. Review service logs and external traffic counters: the voice request should not create connections outside the home network. Repeat the check after another restart.

One local test does not prove that every command or device will work. If a service or model is unavailable, the pipeline should show an error. In Assist mode, the Habab app does not automatically route a failed request through its own cloud path, but other Home Assistant integrations may still connect to the internet.

What is not verified yet

• Russian recognition accuracy with your microphone, voice, and room noise.
• The full STT → local model → action → TTS path with WAN disconnected.
• Peak RAM, CPU, temperature, and latency on a specific server.
• Missing-model, cold-start, and update behavior for the chosen services.
• The absence of external connections, measured during a real voice request.

Treat this as a setup for your own evaluation, not as a ready-made or performance-tested Habab package. Start with one test light and do not expose other entities to the model before checking its behavior and allowed entity list.

Frequently asked questions

Will this setup work after disconnecting the internet?

Check this with the selected services and hardware. Install the images and models first, keep the LAN available, disconnect WAN, and run the complete voice path after restarting the services.

Will Habab Home automatically replace a local service with a cloud one?

In Home Assistant Assist mode, Habab does not switch a failed request to its own cloud route. Another selected integration or Home Assistant pipeline may still connect to the internet.

Which model will be faster?

No target server has been selected and no measurements have been made. Evaluate the models on your server for Russian recognition quality, response time, and peak memory use.

Sources

• Fully local Assist — the standard voice pipeline.
• Wyoming Protocol — connecting local STT and TTS services.
• Ollama — the local conversation agent and limits on Home Assistant control.
• Habab Home installation guide — installation and current distribution contents.