Zro: Uncensored Security Models

Product Introduction

1. Definition: Zro is a private, multi-region inference platform specializing in uncensored, open-source AI models for security research and offensive/defensive workflows. It operates as a technical API and CLI service.
2. Core Value Proposition: Zro exists to provide security researchers, red teams, and developers with unfiltered access to specialized AI models like Dolly 1 Security and Dolly 2 Security, enabling them to bypass the content restrictions of general-purpose models for legitimate security testing, all within a privacy-focused infrastructure that guarantees zero request retention.

Main Features

1. Uncensored Security Models: Zro provides fine-tuned, uncensored variants of leading open models, specifically GLM-5.3 Flash and GLM-5.3, branded as Dolly 1 Security and Dolly 2 Security. These models are engineered to process security-related prompts that typical models refuse, using specialized training data for defensive and offensive security analysis.
2. Privacy-First Inference Infrastructure: The platform operates on a zero request retention policy, meaning user prompts and model responses are not stored after processing. It offers multi-region deployment, allowing users to select inference locations for latency or data sovereignty requirements, as detailed in its regions documentation.
3. Unified API & CLI Toolchain: Zro exposes a standard OpenAI-compatible API (Completions/Responses and Messages endpoints) for easy integration into existing tools. Complementing this is the Zro CLI (npm install -g @moonmath-ai/zro), which provides authentication (zro login) and direct launching of supported coding agents (e.g., zro launch opencode), creating a seamless workflow from terminal to model inference.

Problems Solved

1. Pain Point: General-purpose and heavily censored AI models actively block or refuse to answer queries related to vulnerability research, exploit development, malware analysis, and other critical security testing scenarios, severely hindering legitimate research and defensive hardening.
2. Target Audience: Primary personas include Security Researchers, Red Team Operators, Penetration Testers, Vulnerability Analysts, and DevOps Engineers building security tooling. Secondary users are Developers and AI Engineers needing uncensored model access for specialized applications beyond security.
3. Use Cases: Essential scenarios include: automating the analysis of suspicious code snippets, generating payloads for penetration testing frameworks, simulating adversary tactics for purple team exercises, writing and testing security detection rules, and researching novel attack vectors without ethical gatekeeping interfering with the technical process.

Unique Advantages

1. Strengths & Limitations (Pros & Cons):
   • Pros: Unmatched access to uncensored models purpose-built for security; strong privacy stance with zero retention; developer-friendly via standard API and CLI; transparent pricing and model licensing.
   • Cons: Niche focus limits utility for general AI tasks; reliance on specific open models (GLM series) may not match the raw capability of largest proprietary models; users bear full responsibility for ethical and legal compliance of generated content.
2. Key Alternatives & Differentiation:
   • Major Cloud AI Services (AWS Bedrock, Azure OpenAI): Offer powerful models but with stringent content moderation, making them unsuitable for offensive security prompts. Zro differentiates on uncensored access and privacy policy.
   • Self-Hosting Open Models (via Ollama, vLLM): Provides full control and uncensored output but requires significant technical expertise, hardware investment, and maintenance. Zro differentiates by offering managed, scalable inference without infrastructure overhead.
   • Other Uncensored Model APIs: Few services explicitly cater to security workflows. Zro differentiates through its specific security-tuned model variants (Dolly Security), integrated CLI for agent workflows, and clear focus on the security research persona.

Frequently Asked Questions (FAQ)

1. What are uncensored AI models used for in security? Uncensored AI models are used for legitimate security research tasks such as vulnerability discovery, writing exploit proof-of-concepts, analyzing malware behavior, stress-testing defensive controls, and automating red team simulations that standard models would ethically refuse to assist.
2. How does Zro ensure privacy and prevent misuse of its models? Zro ensures privacy through a strict zero request retention policy and provides clear terms of service. Prevention of misuse is based on user accountability, key controls for API management, and targeting services at verified professionals, as it operates on a trust-but-audit model common in security infrastructure.
3. What is the difference between Dolly 1 Security and Dolly 2 Security models? Dolly 1 Security is the uncensored variant of the GLM-5.3 Flash model, optimized for speed and lower latency. Dolly 2 Security is the uncensored variant of the full GLM-5.3 model, optimized for higher complexity and reasoning depth in security tasks.
4. Can I integrate Zro with my existing security tools and scripts? Yes, Zro is designed for integration via its OpenAI-compatible API, allowing it to be plugged into custom scripts, security orchestration platforms, and tools that support standard chat completion endpoints. Its CLI further facilitates integration with coding agent workflows.
5. Who should use Zro versus a general AI API like OpenAI? Security researchers, red teams, and developers building security applications who hit content refusals on general AI APIs should use Zro. Developers needing broadly capable, content-moderated AI for general coding, content creation, or customer support should use a service like OpenAI.