Lumen.

Loading library…

What the socket error means

Only one usage of each socket address means Ollama is already running on port 11434. Do not start a second copy. The command failed because the first Ollama process already owns that port.

Recommended Windows Desktop setup

Ollama Desktop may run a tray app, a server process, and an updater. The server inherits environment variables from the process that launches it. Changing the user environment in Windows does not update an already-running tray app.

1. Open PowerShell. Do not paste the PS C:\...> prompt text, and do not type Markdown links into commands.
2. Persist the exact deployed origin for future launches:

3. Kill the tray app, server, and updater so the environment is rebuilt:

4. Set the variable in this PowerShell session, then relaunch the Desktop app. Replace the path if Ollama is installed elsewhere:

Do not run a second ollama serve while the Desktop app is running. The Desktop app will spawn the one server that owns port 11434. A bind error means another server is already using that port.

Check that Ollama is working

Open this address on the same computer:

You should see JSON containing installed models. Your result showing TcpTestSucceeded : True means IPv4 is working. The warning for ::1 only says IPv6 localhost is unavailable; it is harmless because Ollama is listening on 127.0.0.1. Run:

The second command should show TcpTestSucceeded : True.

To verify the browser origin specifically from PowerShell, use curl.exe, not curl (PowerShell aliases curl to another command):

The result should be 200. To inspect the allowed-origin header:

Look for Access-Control-Allow-Origin: https://lumendesk.pages.dev. This confirms the running server, not only the registry, has the correct setting.

Connect this website locally

When this website is open on the same computer as Ollama, use:

Do not open Lumen.html by double-clicking it for Ollama. A file:// page sends the browser origin null, so Ollama can return 403 even when its server is healthy. In the Lumen folder, run:

Then open http://localhost:8000/Lumen.html. In AI Providers, choose Ollama, press ↻ live, select an installed model, then press Test.

Use this website from a phone

On a phone, localhost means the phone itself, not your computer. Replace it with your computer's LAN IPv4 address. In PowerShell, type:

Find the section for the connection you are actually using, usually Wireless LAN adapter Wi-Fi. Copy the value beside IPv4 Address, for example 192.168.0.110. Ignore Default Gateway, IPv6 addresses, and disconnected adapters.

1. On the computer, run ipconfig and find the active adapter's IPv4 address, such as 192.168.1.20.
2. Start Ollama so it listens beyond localhost:

3. In this website's Ollama provider, replace the Base URL with http://192.168.0.110:11434, substituting your own IPv4 address if it differs.
4. Allow Ollama or TCP port 11434 through Windows Firewall.
5. Keep the phone and computer on the same Wi-Fi network.

If the browser says CORS or “could not read Ollama”

If the error says allow file://, that is a file-page limitation: the browser actually sends Origin: null. Use the localhost Python server above instead of trying to add file://. This website is now hosted at https://lumendesk.pages.dev; for its deployed version, the exact Ollama origin is https://lumendesk.pages.dev without the /#/home hash. If the error happens on http://localhost:8000 or the deployed HTTPS page, stop the background process with Stop-Process -Name ollama -Force, set OLLAMA_ORIGINS, then start exactly one ollama server and press ↻ live again.

Also check that the URL uses the correct protocol, address, and port. This website must be opened over http://localhost or your normal HTTPS address, not an unreachable computer name.

Model and performance checks

Use a model shown by ollama list. Smaller models start faster and use less memory. After the model list appears in Lumen, select the exact provider/model from the AI model selector before using Chat, summaries, quizzes, or flashcards.

Getting help safely

For setup or debugging, you can ask ChatGPT, GitHub Copilot, Codex, an Antigravity-style coding agent, or another trusted coding assistant to inspect the exact error. Share the command output and environment details, but remove API keys, passwords, private file paths, and personal network details unless they are necessary. Ask the assistant to give one command at a time and to preserve the existing Lumen file rather than rewriting it.

Quick diagnosis

• Bind error: Ollama is already running; do not launch a second server.
• Connection refused: Ollama is not listening at that address or port.
• CORS / could not read: restart Ollama with OLLAMA_ORIGINS configured.
• Phone cannot connect: replace localhost with the computer LAN IP and check Windows Firewall. Only configure a LAN URL when Ollama is intentionally listening on that interface.
• No models: run ollama list and pull at least one model.
• Registry says allowed but browser gets 403: restart the Desktop app from a shell where $env:OLLAMA_ORIGINS is already set.