Hypergraph AI
We secure your network.
Our Mission
See attacks early. Know why. Get ahead.
Hypergraph AI learns how your network normally behaves from the flow data it already produces, and flags anything that deviates from that normal for your security team. No signatures, no sensors, nothing to install.
Why now
Attackers move within minutes, with techniques no rule has seen.
AI is helping attackers scale up and invent new ways in. Signatures and rules only catch what is already known. Defenders need to see what changed, early.
Source: CrowdStrike 2026 Global Threat Report.
The solution
Sees what changed.Explains why.
- Network traffic
- Who talks to whom
- Hypergraph AI
- AI-based Anomaly DetectionDetects what's unusual
- Cognitive AIExplains whyin development
- What changed, and why
- Incidents + their reason
Hypergraph AI reads the flow data you already collect as a graph of who talks to whom. Today it detects what's unusual. Next, Cognitive AI will explain why.
How it works
Hypergraph AI turns the NetFlow/IPFIX flow metadata your network already exports into a graph of who talks to whom, with devices as nodes and their communications as links. Unsupervised graph-based AI then learns how that network normally behaves, as a whole and in groups of devices, without labels or attack signatures, and flags what deviates.
01 · Input
Your flow data is enough.
Hypergraph AI reads the NetFlow/IPFIX flow records your network already exports and your MSSP already collects. No sensors, agents or probes to install. No packet payloads: metadata only.
02 · Model
Sees how devices act together.
Every flow becomes part of a graph of who talks to whom, including interactions among many devices at once. The AI reads each device in the context of the others, so it can spot a group acting in a coordinated new way, even when each device looks normal on its own.
03 · Detect
Learns your normal. Flags what deviates.
Unsupervised AI learns what normal looks like for each network on its own. It needs no labels and no attack signatures: it flags behavior that deviates from that normal, even when no rule describes it. What deviates is shown to your security analysts in a dashboard.
Who it's for
One integration.
Many networks protected.
Hypergraph AI is built to plug into an MSSP once, using the SOC and telemetry it already runs, and reach the many clients it protects. Running your own network? It works the same way, straight from your flow data.
- One integration
- MSSPExisting SOC and telemetry
- Many clients
- Enterprise
- Hospital
- Factory
Research
Built on 7+ years of research
Hypergraph AI comes out of more than 7 years of academic research on graph neural networks for networking at UPC – BarcelonaTech's Barcelona Neural Networking Center, in collaboration with Huawei, Cisco, Telefónica and the Technical University of Munich.
Selected publications
- AutoGraphAD: Unsupervised network anomaly detection using Variational Graph Autoencoders (opens in a new tab)
- PPT-GNN: A Practical Pre-Trained Spatio-Temporal Graph Neural Network for Network Security (opens in a new tab)
- Towards a graph-based foundation model for network traffic analysis (opens in a new tab)
- Deep reinforcement learning meets graph neural networks: exploring a routing optimization use case (opens in a new tab)
- Network Digital Twin: Context, Enabling Technologies, and Opportunities (opens in a new tab)
- RouteNet: Leveraging Graph Neural Networks for network modeling and optimization in SDN (opens in a new tab)
EU research & innovation
European research moves faster when the right partners meet.
Hypergraph AI is actively taking part in proposals for EU-funded research and innovation projects. We are open to working with companies, SMEs, universities, research centres and public institutions that need a specialised partner in AI for network security.
We follow calls under programmes such as Horizon Europe and Digital Europe.
Open to partnering with
- Companies
- SMEs
- Universities
- Research centres
- Public institutions
- Infrastructure operators
Why partner with us
Research roots
Born from more than 7 years of research on graph neural networks at UPC – BarcelonaTech, with peer-reviewed work on graph-based AI for networking and security.
European by design
A European company based in Barcelona, aligned with EU priorities on digital sovereignty, NIS2 and the Cyber Resilience Act.
Easy to pilot
Nothing to install: it works from the flow data that routers, switches and firewalls already export. That makes pilots practical in hospitals, industrial and OT sites, IoT deployments and critical infrastructure.
Our focus
AI-powered cybersecurity
AI that learns how a network normally behaves and flags what changed, without relying on attack signatures.
Graph-based network intelligence
Graph neural networks that read each device in the context of the others, using flow data alone.
OT/IoT security
Visibility for industrial and connected-device networks, where installing agents or sensors is hard.
Critical infrastructure
Early warning for the networks behind hospitals, energy, industry and other essential services.
What we bring to a consortium
Technology provider
Graph-based AI detection for network traffic, adapted to the project's use cases.
Pilots and validation
Deployment and evaluation on real flow data at pilot sites, with no sensors to install.
Research
Contributions on graph neural networks, anomaly detection and network security, building on our peer-reviewed work.
Dissemination and exploitation
Publications, workshops and talks, plus a route to market through managed security providers.
Building a consortium?
If you are preparing a proposal and looking for a specialised cybersecurity partner, let's talk. Tell us about the call, the consortium and the role you have in mind for us.
or email us at info@hypergraph.tech
FAQ
Frequently asked questions
Hypergraph AI is a cybersecurity company based in Barcelona, founded in 2023. We build AI that understands network behavior: it sees what changed in a network so security teams can act early.
Attackers increasingly use AI and move fast once they are inside a network: 29 minutes on average to spread, according to the CrowdStrike 2026 Global Threat Report. Many of their techniques have never been seen by any signature or rule. Hypergraph AI helps teams spot this unusual activity early, without depending on known attack patterns.
Hypergraph AI takes the flow data your network already produces and turns it into a graph of who talks to whom. The AI learns what normal behavior looks like for that network and flags what changed. A second layer that explains why something is unusual, with incidents and their likely root cause, is in development.
Only network flow data (NetFlow or IPFIX), which most MSSPs and organizations already collect. There are no sensors, agents or probes to install on your network or devices.
Yes. Instead of matching known attack signatures, Hypergraph AI learns normal behavior and flags what deviates from it. It needs no labeled attack data, so it can surface new techniques that no rule has been written for yet.
Most network detection and response (NDR) tools look at each device on its own or in pairs. Hypergraph AI models how devices and traffic interact together as a group, and it runs on the flow data you already have. In our October 2026 analysis of 16 NDR vendors based on public sources, none modeled how devices interact as a group, and 11 required sensors or probes.
Managed security service providers (MSSPs), where one integration can reach many of their customers, and any company that wants to protect its network.
No. Hypergraph AI works on flow metadata only, such as which devices communicate, when and how much. It does not inspect packet payloads or the content of your communications.
Since July 2026 we have been running an active proof of concept with NexTReT, one of the largest MSSPs in Spain, on real flow data. The product is not generally available yet, and we are opening early access to MSSPs and any company that wants to protect its network. Managing networks for your company or your clients? Let's talk.
From more than 7 years of research at UPC – BarcelonaTech (Barcelona Neural Networking Center), in collaboration with Huawei, Cisco, Telefónica and the Technical University of Munich. Hypergraph AI also took part in INCIBE Emprende by Wayra (Telefónica).
Still have questions? Let's talk
Contact
Get in touch
Questions, partnerships or an EU proposal? Email us and the team will get back to you.