Paris · France

AI systems. Design. Deployment. Security.

Book a meeting →

Talk about your project

Data processed only to handle your request. Legal notice

What we do Full catalogue →

Audit

AI Audit & Security

Assess the security and robustness of your AI systems before an attacker does.

Attack surface mapping (prompts, tools, data, access)
Offensive testing: prompt injection, exfiltration, jailbreak, agent abuse

Deployment

AI Deployment

Design and ship to production AI systems that work in your real environment.

Internal assistants (HR, support, legal, technical)
Business agents with tool access (CRM, ERP, document stores)

Red Team

AI Red Teaming

Attack your AI systems with the same techniques a real adversary would use.

Attack scenarios calibrated on your architecture
Jailbreak, agent manipulation, exfiltration through tools

Training

Pass AI and AI security skills on to your teams, hands on.

AI dev, AI ops and AI security tracks (1 to 5 days, on site or remote)
Hands-on labs and challenges through x0ne.training

Benchmark

Benchmarking & Evaluation

Measure what your AI systems actually do: quality, cost, latency, robustness · with reproducible protocols.

Multi-model comparisons on your prompts and your data
Agent evaluation: success rate, hallucination, tool drift, cost per task

R&D

R&D & Tooling

Open-source tools maintained by x0ne, integrated into your stack · not a black box.

x0ne harness: safe open-source coding agent, 93 typed actions, TOML policy (MCP server, optional Podman/Incus isolation)
x0ne confidential inference: an inference server that refuses to serve outside an attested enclave (Intel TDX, NVIDIA H100), a signed receipt per request
MCP integration: Claude Code / Codex / Cursor connected to your systems

Why x0ne

Security expertise.

Production, not prototype.

Offensive + defensive.

Fully custom.

Skills transfer.

Offensive security challenge

HackMachina

Attack simulation platform for LLMs, agents and AI pipelines.

Prompt injection, jailbreak, data exfiltration
Agent abuse and tool call manipulation
Multi-model robustness benchmarking
Continuous training for red teams

Explore HackMachina → hackmachina.ai

Open source · Agent infrastructure

x0ne harness

A safe open-source coding agent.
93 typed actions. Declarative policy before execution. Structured results. No raw shell.

Rust

Open source

Local-first

LLM

model

→

action.json

typed action

→

policy engine

check

→

executor

rust / sandboxed

→

result.json

structured result

No raw shell

Agents never run bash directly. Every action is typed, named and declared.

Policy-enforced

Every action is checked against a TOML policy before execution. What the policy does not allow goes to a human confirmation, which denies by default.

Structured results

Results come back as JSON. The model reasons over data, not terminal noise.

Model-agnostic

Anthropic, OpenAI, local models through Ollama, or any OpenAI-compatible endpoint such as vLLM.

Local-first

Runs on your machine. No cloud inference required. No telemetry.

Rust core

Memory-safe. No GC. The executor is small, auditable and deterministic.

Confidential computing · Inference

x0ne confidential inference

Confidential Inference Only.
An inference engine that only runs inside an attested enclave.

Rust

Intel TDX

NVIDIA H100 CC

# early access

contact@x0ne.co

Explore x0ne harness → Explore x0ne confidential inference →

training platform

Organise, track and adapt
hands-on training.

Courses, projects, labs, quizzes, instant feedback and progress tracking. A complete framework for technical training.

See the platform →

Cybersecurity Machine Learning Linux Networking Reverse Engineering Pentest AI Security Low-level Dev

Got a project?

Audit, deployment, training, red teaming.