Working with RFEM
Once RFEM is connected, the assistant works in your open model on your machine. Not on a copy, not on an upload. That is the point of the feature and the reason the rest of this page exists.
Setup is on Connecting RFEM; when something is not working, go to RFEM troubleshooting.
The active model is yours to choose
Dlubal CALC does not switch the active model on its own. Creating, opening or activating a model is your decision, so the assistant asks first rather than acting.
This matters more than it sounds. If it silently created a second model when yours was merely inactive, your work would go into a model you never opened, and you would find out much later.
Example: you ask for a frame while no model is open. You get a question first. Answer it, and the build runs in the model you named.
When the question has a few fixed answers, such as variant A or B, they appear as buttons below the answer. One click answers the question, or you can write your own answer. Either way the reply belongs to the same request and costs no new base price.
A few skills exist specifically to build a model from an external source, such as a photo, a hand drawing or a scan, rather than to edit your current one, precisely so that reconstructing a different structure does not land inside the work you already have open. If your active model already holds an unrelated structure, the assistant asks whether to build the new one there anyway or start a separate model, instead of deciding either way on its own. An empty active model needs no question: there is nothing of yours to protect.
Which changes are permitted
From MCP 1.0.14, an operator can grant separate read, restricted write and administrator permissions. Python tools are disabled by default. Restricted write tools can create a new model and calculate an explicitly identified model. Changes to existing objects and arbitrary scripts require other explicitly permitted capabilities.
CALC asks for your confirmation only where work can be lost: before every deletion, before closing, opening or importing a model, before saving over the model file, and before any step it cannot read. Creating and changing objects, creating a new model and calculating run without a question. Every change to an existing model is listed as a short note in the chat. You can have CALC ask more often in the settings under "Approval before model changes". Identify the target model and intended change. A confirmation permits a call; you still need to inspect its result.
You can decline a proposed correction at any time. Declining does not execute the correction or undo changes that have already been made.
For a read-only diagnosis, ask for the open models, their unique IDs and object counts. This does not create a model or switch to another one.
If you only want a suitable section suggested, the initial comparison should use read access. Explicitly request the intended change when you want CALC to apply the suggestion. A section must not be changed experimentally in the active model and offered for approval only afterward.
RFEM costs 3 credits once per request, and reading is free
All RFEM, RSTAB and RSECTION script calls within a request are included, along with required clarifications, approvals and automatic retries. The surcharge applies only when a script changes or calculates the model. A script that only reads costs nothing: the assistant can look around your model as often as it needs to without that showing up on your bill. Including the answer, a request costs at most 4 credits on Standard or 5 on High. Additional services such as PDF reports are quoted and approved separately. The RFEM surcharge needs no separate confirmation: it is shown in the price line below the input box and on the note for the step that incurs it. No successful script call means no RFEM surcharge. The answer includes a credit breakdown. See Plans and credits.
Check the axis orientation before you trust a coordinate
RFEM models can run with Z up or Z down, and the assistant states which orientation it worked with after it creates or changes geometry:
Global axes: Z points down → column head at z = −4.0 m
Read that line. It is what lets you catch a mirrored model before it reaches a calculation, and a mirrored model produces results that look entirely reasonable.
Watch it happen
Keep RFEM visible while the assistant works. Objects appear as they are created, and if something is being built in the wrong place you see it immediately rather than in the results.
Where to go next
- Creating models for building geometry from a description.
- Running calculations for what to ask for and what comes back.
- Reading results for getting the right load case out.