LiveAgent v1.3.4

LiveAgent v1.3.4

Fixes DeepSeek image-input handling so models with vision support, including the flash family, are no longer treated as text-only, and switches release-note generation to the official DeepSeek API.

Overview

LiveAgent v1.3.4 focuses on DeepSeek provider correctness and release automation. Image input support is now driven by each model's capabilities, so flash-family models are no longer forced into text-only behavior. The release notes generator has also been updated to use the official DeepSeek API with consolidated key sourcing.

Highlights

• DeepSeek models with image-input capability can now accept images; flash-family models are no longer treated as pure text.
• Release-note generation now uses the official DeepSeek API.

Changed

• Updated the DeepSeek provider, native response attachment handling, and provider runtime model factory to respect model capabilities.
• Updated provider settings utilities for input modalities.

Fixed

• Fixed DeepSeek image input handling so flash-family models are no longer treated as text-only (#796).

Internal

• Release notes generation now uses the DeepSeek official API and consolidates secret sourcing (#792).
• Updated provider and settings test coverage, including DeepSeek native, attachments, and input modalities.

Contributors: @coder-hhx