Repository navigation
Grunz: desktop coding agent on abliterated open models #3953
behrnt-slatgng
started this conversation in
Show and tell
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Founder here, so weigh it accordingly.
Grunz is a desktop coding agent for macOS, Windows and Linux. Same shape as a terminal agent — plans a task, edits files across a repo, runs commands, reads output, fixes its own mistakes — except it runs abliterated open-weight models (Qwen, DeepSeek, GLM, Kimi) with the refusal training removed, rather than a frontier model.
The practical reason that matters for agent work isn't edginess: it's that an agent which bails 40 minutes into a refactor because something tripped a classifier is an expensive failure mode.
Honest trade-offs: abliterated variants follow instructions slightly less precisely than their base models, and most hosted open models cap out around 32K context, which an agent loop eats fast. It's a GUI rather than a TUI, so if the terminal is the point, Crush is the better fit.
No account, free credits to start. https://grunzai.com
All reactions