ENNL
Book a call

Tools

Hosted API or a model you run yourself?

Five questions on data, volume, latency, control and skills lead to a recommendation with reasons.

Start the toolAll resources
  • A directionHosted API, self-hosted or both.
  • The reasonsData, volume, latency, control and skills.
  • A next stepWith a decision guide to read.

API or self-hosted model

A direction to start a conversation, not a design. It states no cost and no performance figure: test both sides on your own evaluation set before you decide.

How it works

Three steps to a result

  1. Answer

    Short questions, one at a time, in your browser.

  2. Get the result

    An indication with the reasons behind it and what to do next.

  3. Nothing is stored

    Your answers stay in your browser. Nothing is stored or sent.

About this tool

Five questions about your use case. You get a direction, the reasons behind it and a next step. The points are published, so you can see exactly why.

How the tool decides

Each option adds points to two sides. The side with clearly more points wins, one rule can override the points, and nothing else is used.

  1. Every answer adds points to "hosted API" and to "self-hosted". The table below lists them all.
  2. If self-hosted leads by at least 4 points, the direction is self-hosted. If the hosted API leads by at least 4, it is a hosted API. Anything closer is both.
  3. One rule overrides the points: if the data may not leave your own environment, the direction is self-hosted.
  4. A flag is added when nobody has run model serving and the direction is not a hosted API.

Points per answer

Hosted API, then self-hosted.

  1. 01 / Data sensitivity How sensitive is the data the model will see?

    • Public or low-risk content 3 / 0
    • Internal business information, without personal data 2 / 1
    • Personal data, covered by a processing agreement with the provider 1 / 2
    • Data that may not leave our own environment, because of law, contract or secrecy 0 / 4

  2. 02 / Volume How much work will go through the model?

    • Occasional use or a pilot 3 / 0
    • Steady daily use by one team or process 1 / 1
    • Large, sustained workloads running most of the day 0 / 3

  3. 03 / Latency How fast must an answer arrive, and where?

    • Seconds are fine, or it runs in the background 2 / 0
    • Interactive: a person waits for the answer 1 / 1
    • Real-time, such as voice, or close to a machine or site 0 / 3

  4. 04 / Control How much control over the model do you need?

    • A general model is good enough and provider updates are fine 3 / 0
    • We need to pin versions and know when the model changes 1 / 2
    • We need to fine-tune, inspect or change the model itself 0 / 4

  5. 05 / Team skills Who could run a model for you?

    • Nobody has run model serving yet 3 / 0
    • We run machine learning in production, but not GPU inference 1 / 1
    • A team that runs production machine learning and is on call 0 / 2

Most a side can earn: hosted API 14, self-hosted 16.

API model or self-hosted model: how to choose

Next step

Tell us what you want to build

Bring one process, one platform or one idea. We reply within two working days and say plainly whether we are the right team for it.

Book a callSend a messageor email info@nexops.nl directly

Currently accepting projects for Q4 2026