Every model, dispatched from one workspace

Connect the subscriptions and APIs you already use. Each chat gets the model that fits; hard work goes to strong models, simple work to fast, inexpensive ones; images, video, speech and web search each pick their own source.

Pick a task to see which tier takes it
  • HighHardest tasksClaude CodeOpus 5.5
  • MediumMedium tasksBailian Token Planqwen3.8-max
  • LowSimplest tasksWorkBuddyDeepseek-V4.1-Flash
  • AdvisorAdvises, never actsCodexGPT-6.1-Sol

Illustration: the main model picks the tier for a delegated task by how hard it is; each tier can come from a different source.

All your subscriptions in one place

It runs on your accounts and your quota; Klaus neither meters nor reports usage. When a model is called, the relevant task content goes to the service you chose. Any OpenAI-compatible API works too.

Beyond chat models

Images, video, speech and web search each have one global choice, independent of the model a chat uses.

Use the Codex, Qoder, QwenWork or QoderWork subscription, or a custom API; Codex and custom APIs that support editing can also take a reference image.

CodexQoderQwenWorkQoderWorkCustom API
Recorded in the real client: switch source and model, then set tiers, the advisor, images and video in Settings.

Questions

Does Klaus sell models?

No. Every source uses your own account or key; Klaus does not meter usage or charge for it. Task content goes to the service you chose, and that service's own pricing and quota apply.

Who decides which tier a task goes to?

The chat's main model picks the tier for a delegated task by how hard it is. A tier set to “follow the chat's main model” uses whatever that chat is running.

What if the selected model is unavailable?

A model that is signed out, removed, unauthorised or lacks a needed capability is reported as unavailable; Klaus never quietly switches to another source.