XDA’s piece on giving a home NAS a “local AI brain” hit a nerve this week: the writer ended up with a single dashboard that combined Home Assistant automations, a local LLM chat, and a feed of energy and storage metrics, and the comments thread filled up with people asking what stack they had used. The honest answer is that the “self-hosted home AI dashboard” category is plural; several apps cover overlapping ground and the right one depends on which part of the house you want to automate first. We tested 7 self-hosted apps that run on a desktop or a NAS and combine LLM chat, automation, and home metrics in one place.
What to look for in a self-hosted home AI dashboard
A few features separate the apps that survive a month from the ones that get retired.
- Runs on hardware you already own. The dashboard should run on a Synology, a Mini PC, or a Raspberry Pi 5 without needing a dedicated GPU box (LLM-heavy options excepted).
- Local LLM integration. The point of the “local AI brain” is that the model runs on your machine, not someone else’s.
- Real device control. Lights, locks, climate, energy, cameras. A pretty dashboard with no control is a worse smart display.
- Automation graph. The strong tools combine the LLM step with the rest of the automation graph rather than running it as a separate product.
- Multi-user access and per-user permissions. Family members should not all be admins.
- Observability. Logs, history, and rollback for automations.
Quick comparison
| App | Best for | Local LLM | Pricing | Standout feature |
|---|---|---|---|---|
| Home Assistant | Real device control with AI as a layer | Yes (Voice, Assist) | Free, open-source | Most-supported device integration list anywhere |
| Open WebUI | Multi-user chat that ties into existing tools | Yes (Ollama, OpenAI-compatible) | Free, open-source | Per-chat model switching with permissions |
| AnythingLLM | RAG over your own household docs | Yes | Free for self-host | Workspaces with their own knowledge bases |
| Big-AGI | Polished chat front end with a workbench | Yes | Free, open-source | Beam and split-window for multi-model |
| n8n | The connective tissue between AI and devices | Yes (via nodes) | Source-available | Visual editor that ties LLMs to Home Assistant |
| Node-RED | Flow-based automation favoured by long-time tinkerers | Yes (via nodes) | Free, open-source | The default automation graph for homelab veterans |
| LibreChat | Multi-provider chat with a strong RBAC | Yes (Ollama, local) | Free, open-source | The cleanest team-style chat for households |
The 7 best apps for self-hosted home AI dashboard on desktop
1. Home Assistant — best real device control with AI as a layer
Home Assistant is the anchor of the modern self-hosted home dashboard category. The device integration list (Zigbee, Z-Wave, Matter, Thread, every brand of smart bulb and lock and camera) is the deepest in the space, and the 2024-2025 Assist work brought local-LLM voice control to the same product. Pair the dashboard with a local Ollama instance and you have a single tool that turns “set the bedroom to 21 degrees” into the right Z-Wave call without leaving the house.
The 2025 Home Assistant releases added improved local intent recognition, a redesigned Lovelace editor, and tighter Voice PE puck support.
Where it falls short: The LLM workbench inside Home Assistant is functional, not great. For long-form chat or document RAG, you will pair it with one of the chat apps below.
Pricing: Free, open-source.
Platforms: Windows, macOS, Linux. Container-friendly. NAS-friendly through Docker.
Download: home-assistant.io
Bottom line: Pick this as the foundation of any self-hosted home dashboard.
2. Open WebUI — best multi-user chat that ties into existing tools
Open WebUI is the polished chat front end that pairs with an Ollama or any OpenAI-compatible backend. The multi-user model handles families cleanly: parents get admin, kids get a restricted model list, the dashboard shows usage per account. The function-calling support means an Open WebUI chat can drive a Home Assistant API, query your Plex library, or pull data from a Postgres database with the right tool wiring.
The 2025 Open WebUI release added native MCP support, which opened the door to plug-and-play tool integration with the wider open-source ecosystem.
Where it falls short: It is a chat front end first. For automation graphs and device control, pair it with Home Assistant or n8n.
Pricing: Free, open-source.
Platforms: Docker on any OS. Windows, macOS, Linux for the underlying Ollama.
Download: openwebui.com
Bottom line: Pick this when the household needs a shared, polished LLM chat with permissions.
3. AnythingLLM — best RAG over your own household docs
AnythingLLM is the self-hosted RAG product that turns a folder of household documents (manuals, leases, warranty PDFs, recipes, school newsletters) into a chat-searchable knowledge base. Workspaces split content by topic so the “kitchen workspace” and the “tax workspace” do not contaminate each other.
The 2025 release added agentic features that let the chat call out to tools, which is how a question like “what year was the warranty on the dishwasher?” turns into an answer with a citation.
Where it falls short: Setup is heavier than Open WebUI. The chat UI is functional rather than polished.
Pricing: Free for self-host, Cloud plan for managed setup.
Platforms: Windows, macOS, Linux.
Download: anythingllm.com
Bottom line: Pick this when your dashboard needs to answer questions about your own household documents.
4. Big-AGI — best polished chat front end with a workbench
Big-AGI is the chat front end that takes the workbench idea seriously. Beam lets you ask the same prompt to multiple models and compose the strongest parts of each answer into one. Split-window runs two chats side by side. The interface is one of the most considered in the open-source chat ecosystem.
The local-model integration is good across Ollama, LM Studio, and OpenAI-compatible backends.
Where it falls short: No native device-control integration. Multi-user is supported but not as polished as Open WebUI or LibreChat.
Pricing: Free, open-source.
Platforms: Web (host on any OS). Container-friendly.
Download: big-agi.com
Bottom line: Pick this when the chat experience itself is the thing you care about most.
5. n8n — best connective tissue between AI and devices
n8n belongs in this list as the automation layer that ties LLMs to the rest of the house. The OpenAI, Ollama, and Anthropic nodes pair cleanly with the Home Assistant node, the MQTT node, and the long tail of SaaS connectors. A flow that “summarises the day’s security camera events and posts to a household chat” is a few nodes in n8n.
The 2024 source-available licence change matters for some users; the Activepieces alternative covers similar ground with a permissive licence.
Where it falls short: The licence rule blocks some commercial uses. The canvas can get unwieldy on very deep automations.
Pricing: Source-available, free for typical self-host use.
Platforms: Windows, macOS, Linux.
Download: n8n.io
Bottom line: Pick this when the AI layer needs to drive devices and other apps in a clear visual graph.
6. Node-RED — best flow-based automation for homelab veterans
Node-RED has been the default flow-based automation tool for homelab tinkerers for a decade. The node palette includes Home Assistant integration, MQTT, HTTP, and a growing set of LLM nodes that let you pipe sensor data through a local model and back into a device action. The community-built nodes cover almost anything you might want to wire together.
The 2025 Node-RED release improved the LLM node ecosystem and tightened the Home Assistant companion integration.
Where it falls short: The UI shows its age. The community-built nodes vary in quality. Steeper learning curve for non-developers than n8n.
Pricing: Free, open-source.
Platforms: Windows, macOS, Linux. Container-friendly. Raspberry Pi-friendly.
Download: nodered.org
Bottom line: Pick this when the homelab is already on Node-RED and you want LLM steps inside the existing graph.
7. LibreChat — best multi-provider chat with strong RBAC
LibreChat is the multi-provider chat front end that takes role-based access seriously. The household admin can set per-user model lists, per-user token budgets, and per-user feature flags. The Ollama and local-OpenAI-compatible backends are first-class, and the conversation export is the cleanest in the category for backup hygiene.
The 2025 LibreChat release added MCP support and improved the agent loop for tool calling.
Where it falls short: No native device-control integration. The setup is heavier than Open WebUI.
Pricing: Free, open-source.
Platforms: Docker on any OS.
Download: librechat.ai
Bottom line: Pick this for the cleanest team-style chat where each household member gets their own scope.
How to pick the right one
- If device control is the priority: Home Assistant first, anything else second.
- If shared family chat is the priority: Open WebUI for polish, LibreChat for RBAC.
- If the dashboard needs to answer questions about your documents: AnythingLLM.
- If the chat experience itself is the thing: Big-AGI.
- If the LLM step needs to drive the rest of the house: n8n for the visual graph, Node-RED for the homelab-native feel.
- If the household already runs a homelab orchestration tool: use it as the connective layer and pair it with the chat front end you like best.
Frequently asked questions
Can I run all of these on a NAS?
Most. Home Assistant, Open WebUI, AnythingLLM, Big-AGI, n8n, Node-RED, and LibreChat all ship as Docker containers and run on any modern NAS. The local LLM backend (Ollama, llama.cpp) is the part that wants a GPU; if your NAS is CPU-only, run small quantised models or run the LLM on a separate machine and point the dashboard at it.
Do I need a GPU for a home AI dashboard?
For chat with a large model, yes — or at least an Apple Silicon Mac with enough unified memory. For small models (3B, 7B at low quants) and for the rest of the dashboard (Home Assistant, automation, document RAG over modest folders), CPU is enough.
Which app is the best Home Assistant Voice replacement?
Home Assistant Voice Assist with a local Ollama backend is the most integrated answer. Open WebUI with function calling can replicate the voice flow if you also wire up a wake-word and speech-to-text pipeline.
Can I make the dashboard work without internet?
Yes. Home Assistant, Open WebUI, AnythingLLM, Big-AGI, n8n, Node-RED, and LibreChat all run fully offline if you also host the LLM locally. Cloud-only features (weather APIs, traffic data) need internet, but the core dashboard works without.
What is the lightest stack to start with?
Home Assistant plus Ollama plus Open WebUI is the lightest stack that covers all three layers (device control, local LLM, chat). Add AnythingLLM later when you have documents to index, add n8n when you need cross-app automation.
