Skip to content

Ubiquité v0.1.10

Choose a tag to compare

@edoigtrd edoigtrd released this 12 Nov 23:27
· 3 commits to master since this release

📷 Ubiquité v0.1.10

🚀 Release: Image Search & Reranker Architecture

This update introduces a major new module: built-in image search, fully integrated into Ubiquité.
It’s a fresh feature, and it lays the groundwork for future multimedia capabilities.
As usual, this feature is intended to be as configurable as possible.


✨ New Features

🖼️ Image Search — Brand New

The model can now generate dedicated Searx image queries, fetch results, and pass them through a reranker.
This allow you to search for relevant images based on the conversation context.

🧠 Image Rerankers (Extensible)

Two rerankers are included:

  • VoyageAI Multimodal — embedding-based, great accuracy (requires an API key).
  • Size Reranker — prioritizes larger images, zero external dependencies (and set as default).

Size reranker is intended to be usable as a fallback when no API keys are available ans an example of custom reranker implementation.

The reranker system was designed to be extremely easy to extend.
If you want to implement your own logic, check out backend/infrastructure/rerankers/size.py — it’s short, clear, and a perfect starting point.
Contributions are very welcome.


🛠️ Still in Progress

  • Improving image retrieval speed.
  • More reranking strategies.

Configuration updates

As for configuration, please add this to your config.toml:

[images_search.reranker]
class = "SizeReranker"
[images_search.reranker.config]
metric = "area"

And for voyage

[images_search.reranker]
class = "VoyageReranker"
[images_search.reranker.config]
model = "voyage-multimodal-3"
api_key= "YOUR_VOYAGE_API_KEY"

image