Skip to content

2.4.6 Local models

Latest

Choose a tag to compare

@github-actions github-actions released this 03 Nov 05:22

ParadigmEve 2.4.6 — Better Local AI, Model Control & Recovery

Version 2.4.6 builds on the reliability improvements of 2.4.5, with a focus on making local AI more capable, model selection more predictable, and everyday use more resilient. This release improves how Eve handles long conversations, provider switching, and interrupted workflows.

image

What's new in 2.4.6

  • Better local AI support: Improved Ollama conversations and added independent llama.cpp chat support, with more reliable tool calling and long-running responses.
  • Your model settings, remembered: Thinking mode and context size are now saved per Ollama model. Existing conversations keep their chosen model across restarts and Compact & Resume.
  • EVE-Al v0.3: An expanded 80-point evaluation suite for testing local AI agents, with dedicated working folders and protected grading material. We were able to get a nanbeige model to do up to 32 tool calls in a row, and will be comparing some other ones soon. Go %eveal
  • Smoother setup and recovery: Improved guided setup, manual recovery, and handling of failed connections.
  • More reliable conversations: Better browser recovery, provider switching, and Send/Stop behavior, with additional protection against duplicate messages.
  • Updated runtime: Electron 43.8.6 with Node.js 24.22.1, plus Windows CI stability improvements.

Feedback and testing are welcome! Full Changelog: v2.4.5...v2.4.6