Skip to content
NLEN
Book a call

Your customers want AI. Their rules forbid the models you would want to use.

Soev.ai is the AI engine under your software. Search, agents and evaluation through one API, behind your interface. Compliant, measurable and white-label.

Book a technical callView the platform

A software product with its own interface and an AI feature in it. Underneath that interface is the soev.ai API with the four blocks: data and indexing, search, agents, and evaluation and optimisation.

Sound familiar?

  • Regulation or your customers block the latest external models. "The provider has to be European too."
  • Optimising open-source models and getting them production-ready takes deep AI expertise. And quality is hard to measure.
  • Your own prototype works in the demo but does not reach production. Meanwhile an AI-native competitor overtakes you.

We are your AI engine: search, agents and evaluation through one API, behind your interface. A small open model with our harness performs like a frontier model, at a fraction of the cost, with an audit log of the kind your regulator asks for.

our answer

What you get

[01]

One endpoint for search

The first step is almost always good RAG. Hybrid search with reranking and dynamically generated metadata across your customers' documents, permissions per user.

[02]

Agents in your software

Agents that carry out tasks in your product, with an audit log and human-in-the-loop.

[03]

Evaluation

Every agent tested on real data, with a benchmark you can show your customer.

[04]

Models you are allowed to use

Open models in a Dutch or European cloud, optimised to frontier quality.

One API call

You send the question and the user. You get an answer with sources back, within that user's permissions.

On the roadmap

SDK and CLI are on the roadmap. Want to build with us?

Let us know

Example of a call: a question about a contract, with the knowledge base and the user included. The answer comes back with two sources.

It already works in other people's solutions

around 50x less compute

An open model with our harness beat a frontier model in an AI agent for KPN, with around 50x less energy use and cost.

  • central government

    Ministry of Economic Affairs (MKBot)

    Guides entrepreneurs to the right scheme

    1,900

    Around 1,900 schemes now, ambition more than 10,000; every answer traceable

    In production
  • telecom

    KPN

    AI agent at high volumes

    50x

    Around 50x less compute than a frontier model; built in days

    In production
  • central government

    Ministry of BZK

    Agents that process objections into draft decisions

    Delivered

    Completed
  • municipality

    Municipality of Amsterdam

    AI agent at high volumes

    500,000

    Goal: 25-30% of around 500,000 reports a year

    Phase 1 in progress

A prototype is not production yet.

Sound familiar?

  • Mistakes affect citizens or clients directly, so it has to work demonstrably.
  • It has to run in your own environment and comply with DPIA, BIO and your sector rules.
  • Prototypes without evaluation, an audit trail and operations do not make it to production.

One AI agent that has to work flawlessly in an existing process? We set up the soev.ai blocks for your process and solution, with a benchmark up front and shadow running, and deliver in your cloud.

our answer

Where this is already running

  • Municipality of AmsterdamPhase 1 in progress

    Report assistant for around 500,000 reports per year.

  • Ministry of Economic Affairs (MKBot)In production

    Currently around 1,900 schemes, ambition more than 10,000.

Frequently asked questions

We build it ourselves.

Good. We get your builders to production-ready faster: evaluation, logging and optimisation are already built in.

Will we become dependent on you?

Your data and configurations remain yours. The framework is open source; for the engine you get a right of use.

What does it cost?

We work that out together, based on volume and management. Book a call.

Will our customers allow this?

It runs in a Dutch or European cloud, or in your customer's cloud. Never training on customer data.

Want to know what this means for your product?

Book a technical callEmail us directly

Are you an implementation partner? See the partner page

How a first call goes

  1. [01]
    You tell us what you are running into

    Which systems, which data and which rules.

  2. [02]
    We show how it works for others

    With the measurements to go with it.

  3. [03]
    Together we decide what phase 1 can be

    One use case, with a benchmark as the yardstick.

45 minutes · online or on site · no obligation