• One-click deploy
  • Dedicated GPU
  • No prompt storage
  • EU hosting
  • ISO/IEC 27001 certified providers
  • Blue & Red Team variants

Deploy a private security LLM in one click

Choose a model and a dedicated, EU-hosted GPU is provisioned for you. There is no infrastructure to manage, and your data stays inside your own instance. Use it through a private OpenAI-compatible API.

  1. Start instanceDedicated GPU, EU-hosted
  2. Loading modelWeights loaded into your instance
  3. Health checkThe model answers a test request
  4. ReadyYour private endpoint is live
Illustrative deployment: start instance, load model, health check, ready.
1click
from model to private endpoint
0stored
prompts or outputs kept by a provider
1customer
per GPU, never shared
11models
from compact to frontier scale
€4.99/ h
starting price per hour, billed per started hour

Connect your tools to a private security model

Connect your agentic cyber platform, coding agent or agent harness to a private InferPod endpoint. No migration, no change to your workflow.

Security-native models

Open-weight models adapted to security work by specialised developers, from compact to frontier scale, with tool calling for agentic workflows. Blue Team for defensive work, Red Team for offensive research.

Data stays private

One customer per GPU. Prompts, code and findings never share hardware with another tenant.

Easier approval

Run by a company with an ISO/IEC 27001 certified ISMS, hosted only with EU GPU providers certified to ISO/IEC 27001.

You stay in control

Your own key on an OpenAI-compatible endpoint. Pay per started hour; stopping the instance deletes it and ends the billing.

Security models - ready for real threats

We select leading open-weight models whose weights have been adapted for security work by specialised developers, so they engage with real security work instead of refusing it. Each one runs on a dedicated GPU and is delivered as a private endpoint, in a Blue Team and, where the licence allows it, a Red Team variant.

Built for defensive work that has to engage with attack technique. These models understand active malicious code in depth, including its behaviour, exploit mechanics and intrusion tradecraft, without creating it. Where general-purpose APIs typically refuse or dilute the answer, they respond within defined limits.

  • Incident response
  • SOC analysis
  • Code review
  • Repository analysis

Entry

Compact, efficient models

from €4.99 / hour

excl. VAT

  • Muse Glimmer 30B
  • Gemma 4 31B
  • Qwen3.8 27B

No verification required

Basic

Large open-weight models

from €9.99 / hour

excl. VAT

  • GPT-OSS 120B
  • Nemotron 3.5 Super VL 120B-A12B

No verification required

Pro

For demanding workloads

from €19.99 / hour

excl. VAT

  • GLM 5.3 Flash
  • Qwen3.8 Flash Next
  • MiMo V2.6 Flash RL

No verification required

Frontier

Frontier-scale models

from €29.99 / hour

excl. VAT

  • GLM 5.3
  • DeepSeek V4.1 Flash 1017
  • Kimi K3

No verification required

Discover our models

Shared APIs are not built for security work

Pick your role and see where shared APIs fall short in your daily work, and what changes on your own instance.

Engagement data, tool output and exploit reasoning stay inside your own trust boundary.

Data handling

API LLMClient artefacts such as scan output, configs and credentials are processed by a third party under its retention terms

Private LLMClient artefacts are processed on a dedicated GPU instance, and inference traffic goes straight to your own endpoint

Model behaviour

API LLMProvider-side safety layers refuse or dilute exploit development, payload and tradecraft questions

Private LLMRed Team variants work with offensive technique directly, from vulnerability analysis and exploit reasoning to reverse engineering

Retention

API LLMPrompts and outputs may be logged or retained for abuse monitoring and model improvement

Private LLMPrompts and outputs are never stored, and customer data is not used for training

Use is limited to authorised security work. See the acceptable use policy.

Select. Deploy. Call.

Select, deploy and call your API. Infrastructure complexity stays behind the product.

  1. Select a model

    Choose your variant

  2. Deploy

    Choose a location

  3. Call your API

    Endpoint and key, ready to use

    URL https://de-a1b2c3.example.com/v1
    KEY sk-••••••••••••••••
    • VariantBlue Team
    • LocationGermany

    Illustrative. Works with any OpenAI-compatible client.

Made for sensitive work

Client data, unpublished findings and exploit research need more than a promise. These are the properties every deployment is designed around.

  • Each instance runs on its own dedicated GPU, never shared with other customers
  • Only EU GPU providers certified to ISO/IEC 27001
  • Run by a company with an ISO/IEC 27001 certified ISMS
  • No training on your prompts and outputs
  • No storage of prompts or outputs
  • Your own API key per instance, checked on the instance
  • Traffic goes over TLS straight to your instance, never through our control plane
  • Stopping an instance deletes it completely
  • Data processing agreement and subprocessor list at onboarding

Questions

Is the service available yet?

Not yet. Access opens through a waitlist, and we will contact you as soon as your tier becomes available.

Who is it for?

Security practitioners and teams doing authorised security work: penetration testers, red teams, malware and reverse-engineering researchers, vulnerability researchers, and SOC and incident-response teams.

What is the difference between Blue Team and Red Team variants?

Every model is available in two variants where its licence allows it. The Blue Team variant understands active malicious code in depth and responds where general-purpose APIs typically refuse, without creating that code itself. The Red Team variant has fewer restrictions and is intended for authorised offensive research.

Is verification required?

Blue Team variants need no verification. Red Team variants of Entry and Basic need a verified email, a payment method and a short description of your intended use. Pro and Frontier require full verification with an identity check, company details and a reviewed intended use, and Frontier is available to business customers only. Use remains bound to the terms, the acceptable use policy and applicable law.

Do you filter prompts or outputs?

We do not add provider-side content filtering on top of the model.

Do you store prompts or outputs?

No. Prompts and outputs are never stored, and customer data is not used for training. Inference traffic goes straight to your own instance, and we have no technical access to it.

Which certifications do you hold?

Our company runs an ISO/IEC 27001 certified information security management system, and extending its scope to this service is planned. All EU GPU providers we host with are certified to ISO/IEC 27001. Further certifications such as SOC 2 are listed per provider in the data processing agreement.

Is “EU hosting” the same as GDPR compliance?

No. EU hosting is one building block, and ISO/IEC 27001 and SOC 2 are not GDPR certifications. Compliance also depends on contracts, your own processing and how you use the service. We say what we provide on the security page and in the data processing agreement.

How does billing work?

You pay per started hour of a running instance, counted from the first successful health check. Time before the instance is healthy, and downtime on our side, is not billed. Stopping the instance deletes it. Prices exclude VAT.

Which API do I get?

A private OpenAI-compatible endpoint with your own API key, served over TLS. Existing OpenAI-compatible clients and agent frameworks work by changing the base URL and key.

Request early access

Tell us about your team and the models you want to run. We will contact you as soon as your tier is available.