How do I keep Lumstead’s AI requests on my home network?
Run a downloaded model through Ollama on your home server and select it in Lumstead’s Connections. Choose a model that supports tools if you want it to read saved records. A local Ollama address alone does not prove the model runs locally. Check which model the runtime uses.
By the Lumstead team · Product previewWhat you need
Ollama in a self-hosted deployment
The server owner sets the Ollama endpoint. The default is http://127.0.0.1:11434/v1 on the home server. Family members select an installed model. If Ollama runs on another machine on the home network, use HTTPS with a trusted certificate.
Install a local model and record its exact name.
Download a model that fits the server and supports streamed tool calls. In Connections, select Ollama and the installed model ID. Run the connection check. If the model is missing, check the server’s configured address and the installed model name.
Test a read with a made-up note.
Save a private note called “Hall cupboard” with the text “The blue picnic blanket is on the top shelf.” In a new chat, ask which shelf holds the blanket. Check that the assistant used the saved note. A model can answer ordinary chat questions even when its tools do not work.
Test a change and inspect the saved result.
Ask the assistant to prepare a new note called “Picnic checklist” with the item “Blue blanket.” Review the proposed text before approving it. Open Notes to check what was saved. A chat reply alone does not prove a record changed.
Keep the limits clear.
Local inference covers model requests. A TMDB catalog search and delivery of sign-in codes still use external services. Select a downloaded model rather than an Ollama cloud model if the conversation must stay on your network.
Before you start
Can hosted Lumstead connect directly to Ollama on my home Mac?
No. The private Ollama connection is available in a home deployment. Hosted inference connections require a public HTTPS endpoint.
Why does chat work when the assistant cannot find my note?
Check that your account can see the note and that the selected model supports tool calls. A successful connection check confirms the endpoint is reachable. It does not test whether a specific model can read a record.