What architectural context does AI need?

An office archive holds decisions as well as documents. A detail belongs to a project, material system and revision. A solution that worked in one setting can mislead in another. Giving an assistant access to a folder does not explain those relationships.

Find the detail and its status

Imagine asking for a previous entrance detail. We would want the project, drawing revision, material system and reason for choosing it, with a link to the source. The answer should distinguish an early study from an issued drawing. A fluent summary based on the wrong version gives the next designer a poor starting point.

OpenAI’s file search supports semantic and keyword retrieval. MCP provides a protocol for exposing tools and contextual resources to an AI application. These help with access; the office still defines what belongs in the collection and when a source is applicable.

Start with one reusable task

• Define a small task, such as preparing a facade study from a brief and reference details.
• State the required inputs, permitted operations and evidence to show with the answer.
• Include a case with missing information. The assistant should identify the gap instead of inventing a project fact.

A research direction for Fraktal

Archly is our active visualization product; Falcon is a working research prototype for Grasshopper assistance. Linking architectural methods, reference files and project context in a shared library remains a research direction.

A public library would need permission to share its examples, clear authorship and distinguishable revisions. We would start with one method, an example file and an expected result. It becomes useful when another designer can inspect where an answer came from and decide whether it applies to their project.