Skip to content

1.5.0 - reasoning models; security hardening

Latest
Compare
Choose a tag to compare
@lukemarsden lukemarsden released this 29 Jan 11:28
· 66 commits to main since this release
faebbd7

What's Changed

image

Added support for Deepseek-R1 (llama and qwen variants), with a cute glowing collapsible "thinking" box while the model is reasoning. Unlike OpenAI, with Helix and open source reasoning models you can see what the model is thinking! Also added a feature to rotate OpenAI compatible API keys when used with an external inference provider, for security conscious organizations. Also improves reliability of knowledge scraping from URLs when there are many knowledge sources in the database.

Key features:

Other changes:

Full Changelog: 1.4.18...1.5.0