2026 has seen a wealth of open weight model releases with many being viable options for agentic workloads both at home and at the office. Deploying these at scale, exposing them securely and making them available across the internet has remained a challenge. For those with the compute to run these models there have been few solutions with which to turn those resources into a personal, private inferencing API that is business-ready.
Infersec was released under early-access at the beginning of 2026, promising to address this infrastructure requirement by providing a strong European presence. It now boasts a performance-minded, security-focused scalable pipeline with a wealth of other features that make running your own AI inferencing provider a cinch. You can host your own models, classifiers and inter-connect tool calls (database connectors, file-system adapters, etc.) in your own private network and securely expose them to your AI consumers using Infersec's cloud or self-hosted platforms.
We've been hard at work polishing and testing the platform, adding new features and experimenting with all the latest open models. We're ready to say it's ready, and we're planning to open Infersec to the public on the 1st of December, 2026.
What will Infersec provide on launch day?
Our full platform will be available to subscribers immediately, with trial options:
- 14-day trial licence for the self-hosted option
- €5 credit on the cloud-hosted option
Both platforms boast the same features:
- Extensible, secure inferencing endpoints
- Composable inferencing sources (compute connection)
- Support for NVIDIA, AMD and Apple devices
- Local tool connectors for database and filesystem proxies
- Secure MCP endpoints
- MCP proxy
- Server-side tool calling
- Auditing
- OpenTelemetry connectivity
- Integration with Hugging Face for open weight models
- Automatic model downloading and starting
- Lots more!
The cloud-based instance of Infersec is hosted in The Netherlands, with all data handled by it remaining within the EU.
Do I need it?
You might be thinking "hey, I can probably run that myself!" - and that's entirely fair. There are a great number of decent local LLM engines available, such as llama.cpp, vLLM and SGLang, and you can learn to run such systems yourself on your own hardware. Likely with the help of an LLM of your own. This is part of what we at Infersec provide via our platform, but it's certainly not the entirety of our product.
One machine, one model, one user - llama.cpp will do the job and we won't pretend otherwise. The shape of the problem changes with the second machine, or the first consumer who isn't you.
Imagine you have a fleet of laptops, or desktops, or rack servers, or Mac Minis/Studios - scaling uniformly across this compute is a challenge, especially when you consider that, to support a production-ready inferencing API, you'd also need the following:
- Load balancing
- Graceful onlining/offlining of compute
- Model-fit
- Priority/routing
- Rolling model updates
- Monitoring
Every item there is something you'd otherwise build and babysit yourself. Infersec's Conduit application runs on each machine, reports what it can host, and the control plane handles the rest: which source takes a request, what happens when one drops offline mid-stream, and how a model rolls out across the fleet without cutting anyone off. You keep the hardware - we run the plumbing.
When can I get in?
Public access opens on the 1st of December, 2026. Both trials start that day - the 14-day self-hosted licence and the €5 cloud credit. Pricing has the full numbers, and the docs walk through the rest.