Set up local AI for VerityRead

English · 繁體中文 · 简体中文 · 日本語 · 한국어

VerityRead needs one local AI Provider. Choose either path:

  • Chrome Built-in AI: No separate app is required when your Chrome version and device report the model as available.
  • Ollama: Install it when Chrome Built-in AI is unavailable or when you prefer to manage a specific local model.

VerityRead never switches Providers without your action and does not use cloud inference as a fallback.

1. Install Ollama

Download and start Ollama from the official download page. Chrome Desktop is the supported browser for VerityRead 0.1.0.

2. Install one starter model

Start with one model. You can add or remove models later.

Model Profile Approximate download
gemma4:e4b Quality-first; recommended default 9.6 GB
gemma4:e2b Smaller download for a more limited device 7.2 GB

Quality-first:

ollama pull gemma4:e4b

Smaller download:

ollama pull gemma4:e2b

The sizes above reflect the official Ollama Gemma 4 library checked on 2026-07-26 and may change. Download time and generation speed depend on your network and hardware.

3. Allow only the VerityRead extension origin

Skip this section unless VerityRead reports a CORS error.

Copy the Extension ID shown in VerityRead’s error card or in chrome://extensions. Replace <EXTENSION_ID> below with that exact value. Do not use *.

macOS

Quit Ollama from the menu bar, run:

launchctl setenv OLLAMA_ORIGINS "chrome-extension://<EXTENSION_ID>"

Then reopen Ollama.

Windows

Quit Ollama from the system tray. In PowerShell, run:

setx OLLAMA_ORIGINS "chrome-extension://<EXTENSION_ID>"

Then reopen Ollama from the Start menu.

Linux

If you start Ollama manually:

OLLAMA_ORIGINS="chrome-extension://<EXTENSION_ID>" ollama serve

For a system service, set the same OLLAMA_ORIGINS value in the Ollama service environment and restart the service. See the official Ollama FAQ for environment-variable details.

4. Verify in VerityRead

  1. Open VerityRead → Settings.
  2. Select Ollama.
  3. Select the installed model.
  4. Choose Test Ollama connection.
  5. Continue only after the Provider reports Ready.

The health check reads model metadata from 127.0.0.1:11434; it does not send page content. If Ollama cannot be reached, make sure the app is running. If no model appears, wait for ollama pull to finish and choose Re-check.

Return to the VerityRead introduction