A user with an 8GB RAM laptop successfully ran the Qwen 3.5 4B language model alongside Ollama and Open WebUI. This configuration demonstrated that managing a homelab does not require cloud resources. The setup integrated Docker containers, Home Assistant, backups, and network storage through a single chat-based interface. All system components operated without internet access, ensuring that data remained fully private within the local network.
Through Open WebUI, the user connected a local language model to Home Assistant’s MCP server and a series of n8n workflows. Each n8n workflow was specifically designed for a single task, such as identifying unhealthy Docker containers or confirming the completion of backup jobs. The model determined which tools to use based on the user's input and executed the appropriate workflows automatically.
Automated Task Execution
For example, when a user asked, “Show me which services are unhealthy,” the system initiated a workflow to scan Docker containers for issues. If it found a container that was stopped, the model would then launch a restart workflow and provide the results directly in the same chat. This capability eliminated the need to navigate between different dashboards or command-line tools to manage the homelab.
The user also used the setup for more complex queries. If a media service like Jellyfin was unavailable, the model could first check the container status and then retrieve recent logs to help diagnose the issue. If the container had stopped, it could invoke the restart workflow and confirm the container's new status in the same chat. This integration of Docker monitoring, log retrieval, and workflow execution significantly streamlined troubleshooting and maintenance tasks.
The user prioritized privacy by disabling cloud services in both Ollama and Open WebUI. All key components of the system, including Home Assistant and n8n, were self-hosted. The laptop used only local network connections, ensuring that prompts, device states, and container logs never left the local environment and were never shared with external services.
Privacy and Access Controls
The integration with Home Assistant was carefully configured to limit access. The model could monitor the status of entities that were exposed to it and trigger automations, but it could not access the full Home Assistant dashboard. Users could further restrict permissions to monitoring only, preventing the model from altering device states if desired.
By default, Open WebUI supports external cloud models and search services, but the user disabled these features entirely. Only the local Ollama endpoint and internal services were allowed to operate. This setup ensured that all model interactions and system operations were confined to the local network.
Workflow Organization and Precision
The Qwen 3.5 4B model worked best with clear, well-defined tools. When workflows had similar names or purposes, the model had difficulty choosing the correct one. To avoid errors, the user ensured each workflow had a precise name and focused on a specific task, improving accuracy and reducing confusion.
The system was designed so that the model could link multiple actions, such as checking a service's status and retrieving its logs. However, each step was executed by a separate workflow, which kept the system organized and efficient. This design allowed the model to respond directly to user queries in a structured manner without adding unnecessary complexity.
The model’s ability to combine actions within a conversation was a major benefit. For instance, a request to “Check how much storage remains on the NAS” would trigger a workflow to assess the storage usage. A follow-up action could then check backup completion status. This multi-step interaction within the same interface reduced the need to switch between multiple tools or dashboards.