Devatnet AI

Your organisation's operational knowledge, kept and brought back when it matters.

A private AI grounded in your own material — the decisions, the fixes, the context your team worked out the hard way. Hosted on infrastructure we own, isolated to your organisation, and answerable for what it does.

What it does

Most of what keeps an environment running is never written down anywhere useful. It sits with whoever fixed it last. General-purpose assistants do not close that gap: they are fluent about the world and know nothing about your estate. Devatnet AI is built the other way round — it starts from your material, keeps it, and gets better as your team works.

Retrieval grounded in your estate

Answers are drawn from your own corpus of decisions, incidents, and configurations rather than generic training data.

Tenant isolation enforced in the database

Retrieval is scoped to your organisation at the storage layer, so another customer cannot reach your material and you cannot reach theirs.

Self-hosted model fleet

Inference runs on open-weight models in our datacenter. Routing to a paid third-party model stays off unless it is separately agreed.

Never trained on your data

Your corpus is retrieved from, not learned from. It is exportable on request and deletable on instruction.

Visible consumption

Usage is recorded against your plan as it happens, so what you consume and what remains are things you can see rather than infer.

Works where your team already works

Reachable from the AI tools your engineers already use, or directly over an authenticated API with a credential scoped to your organisation.

Built for

Teams losing the same day twice to problems they have already solved

Organisations that cannot send operational context to a public model provider

Operators carrying knowledge that currently lives with one or two people

Engineering groups who want the answer to improve every time they use it

Plans

You are buying access to your corpus and our model fleet, not tokens. The models are ours, so there is no metered third-party cost to pass on — plans differ by fair-use ceilings and the support that comes with them.

Design Partner

For the first organisations adopting this with us, shaping it as they use it.

500 jobs per month
50,000 retrievals per month
50 GiB corpus

Team

For a working team that has made it part of how they operate.

2,000 jobs per month
250,000 retrievals per month
250 GiB corpus

Enterprise

For organisations running it across several teams and systems.

10,000 jobs per month
1,000,000 retrievals per month
1 TiB corpus

Bring us the environment and the constraint. We’ll work out the right path.

Already onboarded? Your usage against your plan's ceilings is live in the console. Sign in to your console.

Request access