A WhiteRabbitNeo / DeepHat alternative

Short answer: WhiteRabbitNeo is now DeepHat, a cybersecurity model from Kindo. Its older and mid-size models are open weights you can run locally, and if you want that, do it — nothing hosted beats an air-gapped box for sensitive work. If you’d rather not host anything, Obscurify.ai serves cybersecurity-tuned models on an OpenAI-compatible API, on an account that is a random 16-digit number, with no training on your prompts. The newest DeepHat model (Attack v3) is proprietary and runs inside Kindo, not as open weights.

WhiteRabbitNeo was one of the first open cybersecurity model families, trained for offensive and defensive work that general models refuse. In 2025 it rebranded to Deep Hat, now developed by Kindo, and the Hugging Face org points to the same models.

Disclosure: this page is written by Obscurify.ai. Facts about WhiteRabbitNeo / DeepHat are from Hugging Face, deephat.ai and kindo.ai, verified on 2026-11-08; corrections with a source are welcome through the contact page.

What’s open and what isn’t

This matters, because “WhiteRabbitNeo is open weights” is true for some models and not the newest one:

• Open weights (run them anywhere): DeepHat-V1-7B (Apache-2.0, a fine-tune of Qwen2.5-Coder-7B, 128K context), and the older WhiteRabbitNeo 13B, 33B and Llama-3.1 8B / 70B models. Community GGUF builds exist, so you can run them in Ollama or LM Studio.
• Proprietary (hosted only): Deep Hat Attack v3, a 30B Mixture-of-Experts model, is Kindo’s proprietary model and runs inside the Kindo platform [1]. It is the strongest of the line, and it is not an open-weights download.

So the trade is real: the open models you control completely but have to host; the best model is behind Kindo.

Side by side

When to pick WhiteRabbitNeo / DeepHat instead

• You want to run it locally or air-gapped. This is the real reason to choose it. DeepHat-V1-7B and the older WhiteRabbitNeo models are open weights. On a machine with no network, nobody sees your prompts, full stop. No hosted service, including this one, can match that.
• You want one model trained only on security. DeepHat is purpose-built for offensive and defensive cybersecurity, trained on attack patterns and operator tasks.
• You’re an enterprise buying a DevSecOps platform. Kindo runs DeepHat inside a harness with RBAC, policy gates and an audit record, with private execution inside your environment. If that’s the requirement, that’s the product.

When Obscurify.ai fits better

• You don’t want to host a model. No GPU, no weights, no ops. Call a cyber-tuned model over an OpenAI-compatible API, or use it in a browser tab.
• You want a standalone API your own tools can call. Point Cline, OpenCode, a shell wrapper or any OpenAI SDK at https://obscurify.ai/v1. DeepHat has no public standalone API; it runs inside Kindo.
• You want a no-email, no-training account. The account is a random 16-digit number, prompts aren’t stored, and models with the ZDR tag only go to zero-data-retention routes, whose providers’ published policies say they don’t train on or keep the data (published policies, not signed agreements). CyberHeabsy and the Qwen and Gemma cybersecurity models have the tag; the OrcaCyber models don’t, so their provider’s own policy applies.
• You want more than one model. A security engagement is rarely one model: cyber-tuned models for reasoning, coder models for PoCs and tooling, frontier models for the hard parts, all on the same account.

The models

Obscurify.ai’s cybersecurity-tuned models — the OrcaCyber family and CyberHeabsy — are tuned for recon planning, reading scanner and log output, CVE explanation, PoCs for authorized testing, malware analysis and detection engineering. They run on every plan, including Pay As You Go with no monthly fee (no Pro plan needed). The full list is on the cybersecurity LLM page; for coding there’s a separate uncensored coding AI page. For how the broader uncensored models behave, see our open refusal test.

A note on scope

Security here means authorized work only: pentests with permission, CTFs, bug bounty within scope, reverse engineering, malware analysis, detection engineering, OSINT and education. Obscurify.ai’s terms prohibit writing or spreading malware and breaking into systems you aren’t authorized to access. A model that engages with legitimate security questions isn’t a licence to do illegal work.

Start

Guest chat needs no account (30 messages an hour). For the cybersecurity models and the API, an account is a random 16-digit number; add funds with Monero, Bitcoin, Lightning or a card. New accounts get $1 of starter credit, with the Pro models (uncensored ones included) for their first 7 days, while this month’s promo budget lasts.

Sources

1. Deep Hat (deephat.ai): https://www.deephat.ai/
2. Deep Hat web app sign-in: https://app.deephat.ai/
3. Kindo / Deep Hat privacy policy: https://www.deephat.ai/legal/privacy-policy
4. Kindo: https://www.kindo.ai/
5. DeepHat-V1-7B model card: https://huggingface.co/DeepHat/DeepHat-V1-7B

Obscurify.ai facts: pricing, privacy policy, terms, cybersecurity models.

Common questions

Is WhiteRabbitNeo the same as DeepHat?

Yes. WhiteRabbitNeo rebranded to Deep Hat, now developed by Kindo. The older WhiteRabbitNeo models and the open DeepHat-V1-7B are still on Hugging Face.

Is WhiteRabbitNeo / DeepHat open source?

Partly. DeepHat-V1-7B and the older WhiteRabbitNeo models are open weights you can run locally. The newest Deep Hat Attack v3 (a 30B Mixture-of-Experts model) is proprietary and runs inside Kindo.

Can I run WhiteRabbitNeo locally?

Yes, for the open models (V1-7B and the older ones). Community GGUF builds run in Ollama or LM Studio. If you want local and air-gapped, that’s the better route than any hosted service.

Does Obscurify.ai store or train on my prompts?

No. Prompts and replies aren’t stored on the server, and we don’t train on it, and models with the ZDR tag only go to zero-data-retention routes, whose providers’ published policies say they don’t train on or keep the data (published policies, not signed agreements). CyberHeabsy and the Qwen and Gemma cybersecurity models have the tag; the OrcaCyber models don’t, so their provider’s own policy applies.

Can I call it from my own tools?

Yes. The API is OpenAI-compatible at https://obscurify.ai/v1, so Cline, OpenCode, a shell wrapper or any OpenAI SDK works. DeepHat has no public standalone API; it runs inside Kindo.

Last updated 2026-11-08. Plan prices and model lists on this page are read live from the site.