Hindsight

Use Ollama or LM Studio

A local server keeps every screenshot on your Mac, even on macOS 26. Neither Ollama nor LM Studio needs a real API key.

Ollama

  1. Install Ollama, then run ollama pull llava in Terminal. Ollama serves it automatically.
  2. In Settings → AI Engine, choose the OpenAI-compatible engine and use Running a local server? to fill in Ollama's address, http://localhost:11434/v1/.
  3. Leave the API key blank and test the connection.

LM Studio

  1. In LM Studio's search tab, download a vision model, such as llava.
  2. Start the local server from the Developer tab.
  3. In Settings → AI Engine, use Running a local server? to fill in LM Studio's address, http://localhost:1234/v1/, and test the connection.

Connection failed, or blank activities

  • Can't reach the server: make sure the local server is running, not just installed.
  • No models available: pull or load a model on the server first, then test again.
  • The model isn't vision-capable: a text-only model can't read images and returns empty descriptions. Use a vision model such as llava or qwen2-vl.

llama.cpp's server works the same way: enter its OpenAI-compatible base URL.

Still stuck? Email [email protected] and we will help.