Settings

Most settings are preferences. A few change what the answers contain, and those are worth knowing about.

The ones that change answers

Units. The unit system the assistant works in and reports back. Set it once; a mismatch here is a source of results that look wrong by a factor of a thousand.

Under Units & Standards, deflections and displacements have their own Deflection unit and Deflection precision, each before the force settings. Metric defaults are mm and one decimal place; imperial defaults are in and two decimal places. Switching unit systems applies these defaults. The settings apply to new AI answers and existing structured deflection diagrams, including extrema and detail readouts. Lengths and x positions retain their own precision. Previously written answer text remains unchanged.

Design standard. Which design-code family answers use when nothing else says otherwise. The default, Automatic (by region), picks a family from where you work, for example AISC and ACI in the US, CSA in Canada, and Eurocode in Europe. You can also choose Eurocode, DIN, BS, ACI, AISC, or CSA. The setting selects a family only, never an edition or a national annex. A standard named in your message comes first, then your project's design standards, then the standard of the skill in use, then this setting. If the edition, or the national annex of a Eurocode, is still open, Mia asks before calculating: a structure checked to the wrong national annex is checked to the wrong partial factors. Design checks run only on standards Mia can find in the knowledge base. If it cannot, Mia says so and asks for another standard or edition.

Personalization. What you tell Dlubal CALC about yourself and how you want answers written. It shapes wording and level of detail, not the engineering.

Example: you work in a German office and leave the design standard on Automatic. A prompt like "design a 6 m beam for 12 kN/m" is answered under Eurocode, and Mia asks which edition and national annex apply. Name them in your next message, for example "use DIN EN 1993-1-1/NA:2018-12 for the design check", and the check runs with exactly that.

Connection and access

Under MCP servers, RFEM is the only service with an on/off switch. The switch matters because CALC uses it to access a local application and can read or change your model. The row also shows the Dlubal-Connector status and the version details covered in RFEM troubleshooting.

Geo Zone Tool, File Conversion, Dlubal Account Info, Calc Account Info and the Dlubal Knowledge Base are operated by Dlubal. They are available whenever your sign-in and plan permit access, so they have no on/off switch. Available does not mean executed: Mia calls a tool only when your question needs it. A geo-data lookup, for example, costs credits only when it runs.

The connection check waits for your sign-in to load. CALC retries the check once after a temporary error. If it still cannot determine the status, it shows that the check is temporarily unavailable. Sign in again only when the sign-in-required message appears. Use the refresh icon to check the status again later.

If access permissions are not yet known, CALC shows that the connection check is currently unavailable. This does not mean you need a different plan or that the Connector has stopped. CALC checks again when permissions arrive. A confirmed plan restriction is displayed separately.

The RFEM row and the Dlubal Connector menu use the same live connection status. Both show specific causes such as a rejected Dlubal API key. A reachable MCP server alone does not establish a working RFEM connection. The refresh icon starts the shared check again while preserving any required browser permission.

Control the Knowledge Base where the choice applies, in the message bar with Auto, On or Off. With Off, Mia is not given tools for that source.

API keys issues keys for the Developer API, on the plans that include it.

You can also see the time of the fresh connection check and the last recorded Geo call with its outcome and duration. Call history covers up to 24 hours for your current access. Missing history does not establish a connection failure. The display does not start an additional Geo query.

Windows Connector watchdog is a separate switch under Settings → AI assistant. It is off by default and controls the local Windows task that can restart an unreachable Connector every ten minutes. It applies to this Windows user, independently of RFEM access in a chat and Windows login autostart. Internal connection monitoring remains active while the Connector runs. See Connecting RFEM.

Saved RFEM approvals, under Settings → AI assistant, lists everything you kept on an approval card with "This session", "This RFEM model" or "Always for low risk". Each row names the scope, the action, the model, the risk level, the approved credits, the date it was created and when it was last used. You can revoke any approval on its own or reset all of them at once. A saved approval covers only comparable steps with the same risk level and no more than the same credits; deletions and changes with medium or high risk are still confirmed every time. When a step runs without asking because of a saved approval, the chat shows a short line with the risk level and the credits and a link back here.

Conversation context and titles

Conversation history no longer has a fixed setting such as "10 turns". CALC automatically takes the most recent complete messages from the chat and drops the oldest ones first if the selected model's context window would otherwise be exceeded. The stored chat remains complete.

Chat titles are created automatically from the first message. If the short background call fails, CALC uses the first line of the message as a fallback, so the chat does not remain stuck at "New chat". You can rename it at any time.

Example: a long chat contains 30 earlier messages and several large tool results. For the next answer, CALC sends as many of the latest complete messages as fit the model. The older ones remain visible and searchable in the chat.

Language

The interface language, and the language the documentation opens in when you follow a link from the app. Answers follow the language of your conversation rather than this setting: ask in German and you get German, whatever the interface says.

Arabic is available as an interface language. Navigation and dialogs follow the right-to-left reading direction. Code, formulas, and engineering coordinates retain their usual direction. Public FAQs and documentation pages without an Arabic translation open in English.

Choose a Voice Agent

Under Settings → AI assistant → Voice Agent, choose Admin default, Gemini Live or OpenAI GPT-Live, provided the provider is enabled and available. The choice is saved in this browser and applies to new voice conversations. Reconnecting retains the provider selected when that conversation began.

Admin default uses the provider Dlubal CALC sets centrally. OpenAI requires its own consent; an existing Gemini consent does not cover OpenAI. Substantive CALC answers still use the selected Standard/High model. Choosing a voice provider does not change the dictation setting in the message composer.

Dictation

Under Dictation, choose how the microphone in the message composer processes audio:

• Live dictation sends short audio segments to the Dlubal server. Text appears while you speak and the current segment is revised. A pause confirms it without ending the recording.
• Dlubal server · after recording sends the recording after you stop.
• Local on this device processes audio on your device, after downloading a speech model. The quality setting applies to this mode.

The Dictation language defaults to Use CALC language. The expected language follows your CALC interface language, including later changes. You can select a different dictation language or explicitly choose automatic language detection. Specifying a language avoids the extra detection step and can make text available sooner. Previously selected language overrides are preserved.

The microphone dropdown also lets you switch directly between Use CALC language and Detect automatically. This uses the same saved setting as this page and applies to live, server, and local dictation. Stop the current recording before changing it. This selects the expected spoken language; it does not translate your dictation.

Technical correction applies spoken punctuation and your glossary; verbatim mode uses the recognition result directly. The segment pause setting confirms live text, while the separate recording-end setting controls when recording stops. With automatic stopping off, stop manually; recording always ends after at most two minutes. Existing preferences survive the update.

Choose your microphone and use the microphone test to check its input level. This local test ends after ten seconds, does not record audio and does not upload anything. Automatic (recommended) explicitly selects an iPhone/iPad microphone when the browser identifies one. Other devices retain their system route. An explicit selection, such as your headset, takes precedence. Devices are checked again after microphone permission, since labels are often hidden before the grant.

You can edit confirmed text while dictating later segments. Editing the provisional segment ends automatic insertion. After server dictation, optional text polishing sends the draft to the displayed AI provider for a language edit. You can undo the change; review engineering details before sending. Polishing is unavailable in local mode.

Personal speech dictionary

Under Personal speech dictionary, enter one replacement per line, such as Er Wind = RWIND. Use Test dictionary to preview the corrected spelling. Personal entries take precedence over the terms managed centrally by Dlubal and the built-in vocabulary. Verbatim dictation remains unchanged.

The same dictionary applies to Voice Agent. Start a new conversation after changes to supply the current terms as recognition hints. CALC also corrects the complete recognized question before processing it; hints do not guarantee that the native speech transcript always displays the preferred spelling.

Appearance and shortcuts

Theme (light, dark or system) and the keyboard shortcuts list. Nothing here affects a result.

Result diagrams

In the Result diagrams card, the shared Details dialog, and individual diagrams, the settings icon opens Positive curve, Box height, and Show result tooltips. Left and right limits at discontinuities are always shown. These display preferences are saved in this browser and shared by all views. In the result card, box height applies to all four diagrams. Changing the direction affects the drawing only, not the signs of the results. Box height adjusts the vertical display while every x-position stays unchanged. Result tooltips are enabled by default. In compact result plots, they appear as you move the mouse and disappear when you leave. Turn them off across all views for screenshots; extrema and the dialog value readout remain visible. Moving tooltips are excluded from the PDF. Select the calculation position directly in the diagram or through Position x.

Turning tooltips off also hides the moving position line and selection marker. The Copy diagrams camera icon copies a clean image without controls. The scrolling dialog captures all diagrams in full. If the browser does not allow copying images, a PNG file is downloaded. Tooltips switch sides when needed to avoid covering labels. Intermediate values are marked linearly interpolated; the pointer snaps gently to calculation points. All plots in the scrolling dialog share the same x-position.

Where to go next

• Plans and credits for what your plan includes.
• Connecting RFEM for the connector and the API key.