Ollama's New Update Lets AI Search Inside Images and Audio
Now you can search your photos and voice notes just by typing.
Deep Dive
The article reports that the EmbeddingGemma2Model architecture has been implemented on the MLX runner: a 24-layer bidirectional text encoder with PLE, along with shared gemma4 vision and audio towers, producing mean-pool and L2 output. According to the article, /api/embed accepts per-item media via input dicts.
Key Points
- Ollama's new test version lets you search images and audio using text, all on your own computer.
- Everything stays local, so your private files never go to the cloud.
- It's still experimental and requires technical skills to set up.
Why It Matters
This could let you privately search your photos and voice notes without uploading them to Big Tech.