Hyperfences

Hyperfences

Hyperfences are rules that define what your LLM can and can't discuss. Switch on the built-in ones, or build your own based on topics specific to your business.

Built-in rules

Free accounts get the core security rules enabled by default.
Pro unlocks the full rule set, along with custom Hyperfence training.

Security
  • Prompt Injection & Jailbreak

    Free

    Blocks attempts to override system instructions, hijack prompts, or bypass model safety guidelines.

Content
  • Sexually Explicit Content

    Free

    Blocks sexually explicit or adult content in both prompts and responses.

  • Profanity & Offensive Language

    Roadmap

    Filters crude or offensive language in both messages and responses.

  • Hate Speech

    Roadmap

    Blocks discriminatory or hateful content targeting individuals or groups.

Custom
  • Create your own Hyperfence

    Pro

    Define custom topic-based rules tailored to your use case. Block or flag specific subjects, terminology, or conversation patterns unique to your business.

Block vs Allow

Block Hyperfences

Content matching the rule is stopped. Use these to keep specific topics out - violence, competitor mentions, legal advice, and so on.

Allow Hyperfences

Content must match at least one allow rule to pass through. Use these to keep your chatbot focused on a specific domain - your product, your industry, your use case.

Custom Hyperfences Pro

Train your own rules, with no ML knowledge needed. Give Hyperfence examples of the type of content that you want to block or allow, and it builds the hyperfence automatically.

  1. 1 Provide examples of the content that you want to detect.
  2. 2 The hyperfence is built from your content automatically.
  3. 3 Switch it on.

Rules work across languages. A Hyperfence trained on English examples will also catch the same content in French, Spanish, Mandarin, and others. Rephrasing does not help either; the rule matches meaning, not exact words.