Chat basics

Steinkauz AI chat lets you talk to models through the policy layer: you choose a model available on a configured provider, send messages, and get streaming replies. You can switch models at any time within a conversation or start a new one.

Everything you do in chat runs in your active organization. Switch in the organization switcher to work in a different organization. See Organizations.

Chat enforces data routing policy before model egress.

Starting a conversation

Open the chat interface from the main navigation.
Confirm the correct organization is active in the organization switcher.
Optionally pick a model from the model selector. The list shows models returned by the selected provider.
Type your message and send. Replies stream in as they are generated.

Stopping and switching conversations

Replies continue generating on the server until they finish or you stop them.

Leaving a conversation (switching to another chat, opening a new one, or navigating away) only disconnects your view. Generation keeps running in the background, so you can browse other sessions without canceling the reply.
Stop generating cancels the current reply. Use this when you want to halt the model mid-response.
If you return to a conversation that is still generating, chat can resume the live stream so you see the reply as it continues.

Multiple models in one place

You are not limited to a single provider. You can use a Claude model for one reply and an OpenAI or other model for the next. Each message can use the model that is currently selected.

Data routing and the shield control

Steinkauz AI enforces data routing policy on every send. The shield next to Share in the chat header shows the conversation’s effective data sensitivity and lets you choose Auto or Manual sensitivity mode:

Auto: The session floor follows your context default and can rise when you attach sensitive files or select providers that require a higher floor.
Manual: You fix a sensitivity level; only providers allowed by your routing matrix for that level appear in the model picker.

If a send would violate the routing matrix, Steinkauz AI blocks it before inference and explains how to fix it.

Configure the routing matrix, default sensitivity, and provider environments in Settings → Routing policy and Settings → Providers while the relevant organization is active. See Data routing policy.

Your conversations are saved in your active organization. Each member sees only their own threads in the sidebar. Owners and admins do not get a shared inbox of other members’ chats; they use Usage & Activity for oversight.

To rename a conversation, open the conversation menu in the sidebar and choose Edit title. You can set the title (up to 50 characters) and emoji. New conversations may also receive an automatic title depending on your organization’s Title generation settings.

To let someone view a conversation without Steinkauz AI access, use Conversation sharing to create a shareable link.

Usage and costs

Usage is billed by your providers (BYOK). Optional Budgets can cap spend per member or API key. See Usage, costs, and activity details.

Assistant replies expose quick usage metadata and a link to full activity details from the message actions menu. View all activity for a conversation from the session activity panel in the chat top bar.