short answers to fair questions.

kint is an agentic decision intelligence platform. We set out to make optimization accessible through AI agents. Working with customers, we saw that the model is rarely the hardest part. Teams struggle to prepare their data, to write down rules and constraints, and to meet legal requirements. kint handles these steps together. The decision agent prepares the data with the team and captures the rules in the planners' own words. kint's forward deployed engineers (FDEs), who work directly with the customer, step in where a decision needs them. The platform builds models from rules, learns them from data where rules are missing, or combines both. For optimization models, the solver states its bound and gap: how far from best the plan can be. Every result keeps its record: data version, model version and who released it. Drafts of the documentation that regulations such as the EU AI Act ask for come with it. kint runs under four industry flagships: manifest for ports and logistics, merit for energy and grids, takt for manufacturing, and horizon for ERP planning.

Recurring planning decisions under hard limits: berth allocation, equipment per shift, crew rostering, redispatch, maintenance windows, production schedules, stock levels and asset renewal. If your team compares options in a spreadsheet and has to defend the choice later, kint can usually help. The problem types page shows the common shapes.

No. kint sits above the systems you already run, such as the TOS, the ERP, a planning suite, a VPP or the MES. Data comes in as read-only exports. Approved plans go back only through an API you control; the control system decides whether to apply them.

No. A forecast tells you what might happen. kint tells you what to do, within your rules, and how close that plan is to the best one possible. Forecasts can be an input: kint uses machine learning where demand or workload has to be estimated, and then optimizes on top of it.

No. Planners describe the goal and the rules in their own words. kint's decision agent writes them down as a model. In the pilot, kint's FDEs review the model with domain experts. The planning team keeps the decision and gets a bound and a gap for the plan it stands behind.

Your team. The decision agent proposes, and a person on your team releases each plan before it is written back. An approved workflow runs whenever your systems call it. The release is part of the record, together with the data and model versions behind the plan.

Read-only exports from the systems that hold the facts of the decision: the ERP, the TOS, an asset register or a spreadsheet. Every upload is versioned, so each result can be traced to the exact data it used.

Yes. For systems known only from data, kint trains surrogate models on past data and optimizes over them. This is blackbox optimization. The result states which data the surrogate learned from, so your team can judge how far to trust it. See blackbox optimization.

No better plan exists for the model as stated, within the solver's numerical tolerance. The solver keeps a bound that no plan can beat and reports the gap between that bound and the best plan found. When the two meet, the gap is 0.0 %. If a time limit stops the search early, the remaining gap says how much could still be gained. Bound and gap are stored with the run.

Language models approximate. They cannot guarantee that every rule holds or that the plan is optimal, and two runs can give different answers. kint uses a language model for what it is good at: understanding the description and drafting the model. Solvers do the solving. Established open-source solvers are included; commercial solvers are supported, with licences arranged per engagement.

The rule that decides the outcome. Most rules in a plan have slack. A few are met exactly, and loosening one of them would improve the result. kint names those limits in plain language and states what one more unit would be worth. That is where a negotiation or an investment pays off.

Yes. Each run pins the data version, the model version and the solver settings, so the plan can be re-checked against the record. You can compare it with what was decided at the time, or rerun the same model on newer data.

Yes. Each scenario is a versioned variant of the same model and data: higher demand, a crane out, a smaller budget. All variants run on the same engine and can be compared before a plan is released.

Yes, per engagement. kint connects your own GPU servers or cloud GPUs and runs GPU-capable solvers on them. GPUs pay off for very large linear programs and for routing problems. Many mixed-integer models solve just as fast on CPUs. So kint measures the models on both before anyone plans hardware.

There are three options. kint is hosted in Germany and operates the platform itself. For operations that must not leave your network, we scope an installation inside it. Software vendors embed kint in existing planning software through the API. For each setup, the documentation lists which components process the data and where, including the language model that drafts models. See security.

kint processes the data under the GDPR. The data is not shared with other customers, and kint trains no models for other organisations on it.

The act sets duties for record keeping, transparency and human oversight, and they are part of how kint runs. Each result is logged with its data and model versions as an audit trail. It is explained in plain language and released by a person on your team. Drafts of the technical documentation and a risk register are kept current from the running system, for your compliance team to review.

A fair question. Plans carry their bound, gap and record, so your own people can check them. Data and records can be exported as JSON and CSV. The about page introduces the team.

With one decision. A workshop with the planners scopes one recurring decision: goal, rules and data. In a pilot, kint builds the model with the team and compares it with today's plans. An approved workflow then runs whenever your systems call it. A person on your team releases each plan before it is written back.

It depends on the setup. Options are a pilot on one decision, the operated platform, an installation in your data centre, or an OEM licence for software vendors. The pricing page explains how engagements are structured.

Data and records can be exported as JSON and CSV; model export is agreed per engagement. kint is hosted in Germany and operates the platform itself.

Yes. Vendors of planning, ERP, TOS or energy software can embed the kint engine as an OEM component. Their product then offers plans with a bound and a gap. See oem and api.