1StopTek

AI infrastructure and developer platform

1st AI Platform

1st AI Platform is the AI infrastructure and developer platform operated by 1StopTek. It provides AI processing on infrastructure we operate ourselves, through one versioned API. Eurylie, our personal assistant, is built on it.

Developer access is in testing

The API, developer console and metering are being verified in staging. The public API is not open yet, no production keys are issued and nothing is charged. This page describes what is implemented today and labels everything else.

Overview

One platform, two ways to use it

Eurylie is powered by the 1st AI Platform, with shared safety protections and centrally managed AI services.

Eurylie

Our personal-assistant application. It uses the platform's conversation, document and memory features through its own app interface.

Developer API

A versioned HTTP API for your own products: chat completions, streaming, structured output, embeddings and usage, with project-scoped API keys.

How a request travels
  1. Your application
  2. 1st AI Platform
  3. Safety Layer
  4. AI processing

Self-operated inference

Models run on hardware 1StopTek operates. The API does not forward your requests to an external AI model provider.

Metadata-only metering

Usage records hold ids, model, token counts, status and latency. They never contain prompts or outputs.

Limits you control

Spending alerts, hard spending limits, rate limits and monthly token allowances are enforced by the platform before a model is called.

Use cases

What teams build with it

Typical uses of the capabilities in testing today. They describe what the API supports, not finished customer deployments.

Assistants inside your product

Answer questions and draft replies in your own interface, with your own system instructions.

Structured extraction

Turn free text into JSON that matches your schema, validated before it reaches you.

Search and classification

Embed documents and queries to rank, group or route content.

Data that stays in-house

Use AI where requests must not go to an external AI provider.

Website assistants

Website assistants

A website assistant answers your visitors' questions from your own website's content and suggests links to your pages. You set it up in the console in six guided steps and add one line of code to your website.

Example only: website assistants aren't available to customers yet.

Answers from your content

It answers questions from the knowledge you add: pasted text, .txt or .md files, or up to 30 pages imported from your confirmed website. It suggests links to your pages; it doesn't take orders or bookings. Word and PDF files aren't supported yet.

Only on your website

You confirm that you own the website before the assistant can go live, and it only answers requests coming from that website. Visitor and spending limits apply as well. The code you add contains no key.

A spending limit first

Nothing goes live without a monthly spending limit for your organization, and the assistant stops answering when the limit is reached.

Setup in six steps

  1. 1Add website
  2. 2Add knowledge
  3. 3Personalise
  4. 4Preview
  5. 5Install
  6. 6Go live

Capabilities

Capabilities

Every capability is labelled with its current state. We only publish examples for endpoints that are implemented and tested.

Capabilities
CapabilityStatus
Chat completionsIn testing
StreamingIn testing
Structured outputIn testing
EmbeddingsIn testing
Model discoveryIn testing
Usage APIIn testing
Document retrieval (RAG)Planned
Tool and function callingPlanned
Image and audio inputNot supported

Safety Layer

Inputs and outputs pass the platform's Safety Layer. A refused request returns safety_error; a refused output ends with finish_reason content_filter.

Pricing

Pricing

Usage-based, per token: you pay for the input and output tokens your requests use. Testing stays free: nothing is charged until paid access opens and your organization accepts the paid terms.

Rate card rate-2026-11-12-r4

Prices

Billing unitMeasured asPrice per 1M tokens
Chat input tokensprompt tokens per request, same price for every model€0.188
Chat output tokenscompletion tokens per request, same price for every model€1.572
Embedding input tokensinput tokens per requestNot included; priced separately
Requestsnot billed separately—
Document processing, storagenot offered on the Developer API—

What counts as billable usage

  • Input tokens are the prompt tokens the model processes for your request; output tokens are the tokens it returns. The console shows both for every request.
  • Embeddings are not included in the token prices above. Their cost is separate and is announced before paid access opens.
  • Requests rejected before they reach a model, for example because of an invalid key or an exceeded limit, use no tokens and cost nothing.
  • Thinking, tool use, context caching and batch processing are not offered, so they have no separate prices.
  • A change of price is published as a new rate card version before it applies.

Commercial terms

Currency
Prices are in euro per 1 million tokens, before VAT. The exact cost of each request is kept without rounding; amounts are rounded once, to the cent, on the monthly statement.
Who can buy
Paid access will first open to business customers established in Germany, with German VAT added to the net. Everyone else can keep testing free of charge; paid terms for other customers follow once their tax treatment is confirmed.
Included allowance
Announced before paid access opens.
Minimum charge
None. You pay only for the tokens you use.
When usage becomes billable
Only after paid access opens and your organization accepts the paid terms. Usage during testing is never billed retroactively.
Limits
Tier limits apply in addition to any spending limit you set.

Security, data processing and support

Security, data processing and support

What we can state today, based on the implementation, and what is still being finalised.

Where it runs

Models run on inference hardware operated by 1StopTek. The hosting location of each environment is not confirmed yet; it will be listed here once it is, before public launch.

No external model provider

Your requests are not sent to OpenAI, Anthropic, Google or any other external AI provider. Inference stays on hardware we operate.

Content

The Developer API does not store request or response content. Usage and audit records contain metadata only. Requests are processed by the Safety Layer, which keeps pseudonymised safety events for 30 days.

Services we use

Account emails are delivered through our email provider. TLS certificates come from Let's Encrypt. Neither receives your API content. Any further processor will be listed here before it is used.

Retention

Proposed: usage records 24 months, audit records 13 months. Final periods are set with our data-protection review before launch.

Support and disclosure

Developer questions: developers@1stoptek.com. Include the request_id from the error. Security reports: see our security page.

Building on the API?

Quick start, authentication, models, code samples, limits, errors and the API reference are in the Developers section.