2026.9.1

Heads-up after upgrading: open Settings → AI once. Some model choices get reset to the new defaults:

• models OpenAI or Anthropic have retired, or will retire on Oct 23 (such as o3-mini)
• the original default models, if you never changed them

Pick a different model there if you prefer. The built-in OCR instruction is also replaced; anything you added in the custom field is kept.

AI models

• New defaults: gpt-6-luna (OpenAI) and claude-sonnet-5-5 (Anthropic). The OpenAI model list now includes the GPT-6 models.
• If a saved model is no longer available, Settings → AI switches it to one that is. AI features now name the missing model instead of showing a generic error.
• Clearer message when Claude declines a request. The auto-tag button now shows the actual error.

OCR

• Fixed: OCR on new installs had no instruction. Models described the picture instead of copying its text, and Claude rejected the request outright.
• Images are converted before sending when needed (GIF and other formats, very large photos, sideways phone photos), so one image works with every provider. HEIC images now get a clear message.

Ollama

• On Ollama 0.31.2 and later, models skip the thinking step, so short tasks don't wait on a long reasoning pass.
• Embedding models no longer appear in the model list, and Ollama cloud models are marked "(cloud)".
• Ollama's own error messages are shown, for example when a text-only model is asked to read an image.
• Settings and the README no longer say notes never leave your server, because Ollama's cloud models send them to ollama.com.

Privacy

• OpenAI requests now ask OpenAI not to store them.

Under the hood

• The Docker images now use Node 22, with updated OpenAI and Anthropic libraries.

image: ghcr.io/alexmicuplusfour/itsnotes-server:2026.9.1
image: ghcr.io/alexmicuplusfour/itsnotes-client:2026.9.1