Console
API/Getting Started/Introduction

AdaptiqAI Developer Platform

Use transparent open-weight models through an OpenAI-compatible API, enhanced with persistent Native Memory and knowledge your application can control.

Early access
We give access gradually, so everything runs smoothly and stays stable as more people join. If you can't get access yet, you're welcome to join the waiting list. Join the waiting list
Up to 5M-token context in each agent session
Every request runs with Native Context Memory. Each agent session holds up to 5M-token context, plus Native Knowledge. Memory works across your conversations, agents and tools, is kept per workspace and per model, and stays between sessions. Your workspace is isolated, so other customers can't reach your data. See pricing.

What you get

  • OpenAI-compatible — use the official OpenAI SDK or any compatible client
  • Native Memory — persistent context without replaying the full history in every request
  • Native Knowledge — add sources in the Console and enable them where needed
  • Model transparency — the underlying base model and runtime are visible
  • Runs in European data centres — operated by AdaptiqAI, a Danish company
  • Pay per token — DeepSeek V4.1 Flash 1017 $0.60 input, $2.00 output and $0 kv cache per 1M tokens. You pay from a prepaid balance and add funds in the Console. Prices in USD, excl. VAT. VAT is added at checkout based on your country.

How a call works

Seven steps, no new SDK.

  1. The client sends an ordinary OpenAI-compatible request — typically with the whole history, as it always has.
  2. The provider identifies workspace, user and conversation. Identity comes from the key and the call’s metadata, not from the prompt text.
  3. The submitted history is compared with what already exists in the bounded timeline. Everything recognised is dropped.
  4. Activated Native Knowledge is available to the model directly, not as text inserted into the prompt.
  5. The model processes only the new input.
  6. New messages, tool calls and answers are registered in the timeline.
  7. The response is returned with an accounting of logical context, new input processed, previous context re-processed, and output — see what comes back.

Where the memory lives

Outside VRAM. Active contexts stay ready, held where an answer can start immediately. Inactive ones move to cheaper storage and are fetched back when the user returns.

That is what sets the ceiling. In an ordinary inference stack, KV-cache reserves VRAM per user — and that, not compute, limits concurrency. With the context outside VRAM, the ceiling is throughput instead, so the same hardware carries markedly more.

Get started

API overview

The API follows the OpenAI request format. Use Authorization: Bearer … and discover the current model catalog through GET /v1/models.

EndpointPurpose
POST /v1/chat/completionsChat and agentic model calls
GET /v1/modelsCurrent models and capabilities

Support

Questions? Write to support@adaptiqai.com.

Was this page helpful?