Heretic removes restrictions from language models
- ID
- 27035
- Status
- summarized
- Published
- 21 Sep 2026, 12:35 PM
- Fetched
- 22 Sep 2026, 8:17 PM
- Provider
- Hacker News
- Category
- dev-community
- Original URL
- https://heretic-project.org/
- Source URL
- https://hnrss.org/best
Summary
- Score
- 5.5
- Created
- 22 Sep 2026, 9:25 PM
- Tags
- Audience
- developersai_ml_learnersai_agent_users
What happened
Heretic is a pip-installable tool (heretic-llm) that strips safety/restriction layers from local language models so they follow all instructions without refusals. You can run it with a single command like `heretic Qwen/Qwen3.5-4B`, and it's AGPL v3+ licensed, created by Philipp Emanuel Weidmann and contributors.
Why it matters
If you're building agents or pipelines on local open-weight models and hitting refusal guardrails that break your workflows, Heretic offers a one-command way to remove those restrictions. However, the landing page provides zero technical detail on how it works, no benchmarks, and no independent validation—evaluate it on a throwaway model before relying on it, and note the AGPL license implications if you bundle it into a product.
Discussion angle
Is removing safety guardrails from local models a practical necessity for agent builders, or does it create liability and quality issues you wouldn't get with a properly prompted model? Try it live against a small Qwen model and see if outputs actually improve or just become less filtered.