Qonic has introduced AI agents that can find, check and update model elements from inside its BIM platform, alongside an MCP server for connecting external AI assistants. Qonic describes its platform as cloud-based and collaborative; the agents work inside it rather than alongside it. The announcement gives no supported-version list. Users can move from asking about model data to asking the software to act on it—whether that means filling a property, setting up a report or editing geometry.
Qonic says its agents use the platform’s own tools. Prompts described in the announcement include isolating and colouring furniture by level, identifying windows without a fire-rating value, and assigning a value such as F60. Users can also ask for a report to be grouped by building and concrete type, then adjust its settings manually or through another prompt.
Prompts can change geometry as well as data
The agents can also edit geometry. Qonic’s examples include resizing selected windows and creating a building or facade from prompts and reference images. The company says users can choose between Sonnet, Opus and Astra for AI requests, depending on how much reasoning a task requires; the announcement does not specify how model choice affects accuracy or the reliability of particular operations.
I would treat geometry-changing prompts as proposed edits. Qonic says Qonic AI can only change models the user is permitted to edit, and its changes can be undone in the same way as a manual operation. In Qonic’s example, selected windows are widened by 0.5 metres “to the right,” with the direction interpreted from the user’s viewpoint. That means a viewpoint-dependent “right” may widen the wrong edge. For a request like this, I would verify that the intended windows are selected and that each 0.5-metre change is on the intended edge; if either check fails, undo the edit before proceeding.
For fire ratings, I would not fill every missing value with F60 simply because the value is absent; a missing rating does not establish that F60 is the correct requirement. Assign F60 only when the project’s fire specification confirms it for each selected window; otherwise, verify the requirement before changing the model.
MCP connects an external assistant to Qonic
The MCP server offers another route into the workflow. Qonic says external assistants—including Claude, Codex and Mistral Le Chat—can use it to look up projects and models, query elements, calculate quantities, export to IFC, change elements and manage library items. The announcement does not specify whether MCP changes are published automatically or describe role-based limits for those actions.
The two routes also have different costs and setup requirements. Ask Qonic runs on tokens: paid plans include a monthly allowance, and users on the free plan can buy a one-off token pack starting at €50. Purchased tokens do not expire. MCP requires connecting an external assistant; the announcement does not state the cost of using that assistant.
Qonic’s announcement describes a broad set of possible actions, but does not establish how well prompts perform across varied project models or complex edits. A sensible first evaluation is a familiar model and a bounded task: check whether a known set of elements has a required property, inspect the selection and result, then try a reversible edit. That is a way to assess whether the workflow saves effort without treating a fluent response as proof that the model is correct.
Sources
- Qonic November Release Notes 2026 — Qonic
- Qonic AI launches inside BIM platform — AEC Magazine
