If your security team won't let candidate data reach a vendor's cloud, most screening tools have no answer for you. This is ours: the same product, deployed on your infrastructure, screening with a model that runs there too.
This isn't shipping yet. The hosted product is live today; the self-hosted deployment described below is what we're building next, with our first customers. We'd rather tell you that up front than after a signature.
Screening runs on a model inside your network — DeepSeek, GLM, or whatever your team already runs — over a standard OpenAI-compatible endpoint. No resume text is sent to a model vendor.
Your Postgres, your backups, your retention policy. Candidate data lives where your other candidate data lives.
No telemetry, no analytics, no licence check calling our servers. Point a packet capture at it — that's a reasonable thing to do, and we'd expect you to.
A container on your servers, behind your reverse proxy and your TLS, upgraded when you decide to upgrade it.
You'll find this out in a security review anyway. Better you hear it from us first.
In our hosted product, the agent can research a company on the live web. That feature is a search engine, and a search engine is the internet — so on an isolated deployment it's switched off. You'd supply that context yourself. We'd rather cut the feature than quietly route your data out to keep it.
We're building this with our first self-hosted customers rather than shipping a box and wishing them luck. That means a conversation about your environment, your model, and your timeline before anyone signs anything. It also means it costs more than the hosted product — a dedicated deployment is dedicated work.
Screening quality depends on the model doing the reading, and an open model running on your hardware is not the same as the hosted one. We measure that against your actual roles and resumes during scoping — and we'll tell you what we find, including if it isn't good enough yet.
What you run, what your security team needs, and how many roles you screen. If it's not a fit, we'll say so.
Talk to us