--- title: "The Fremtidens Tegnestue toolbox" canonical_url: https://fremtidenstegnestue.dk/en/toolbox markdown_url: https://fremtidenstegnestue.dk/en/toolbox.md author: "Fremtidens Tegnestue" page_type: static primary_intent: "Information about buildings is spread across articles, guidance and reports. We use artificial intelligence (AI) to find and bring together relevant information from five Danish publishers. You receive an overview with links to the original texts and questions to investigate further." primary_keyword: "building knowledge toolbox for AI agents" audience: "architecture studios, municipalities, foundations" summary: "Information about buildings is spread across articles, guidance and reports. We use artificial intelligence (AI) to find and bring together relevant information from five Danish publishers. You receive an overview with links to the original texts and questions to investigate further." source_basis: - "Bevica: https://www.bevicafonden.dk/" - "Bolius: https://www.bolius.dk/" - "Realdania: https://realdania.dk/" - "Grundejernes Investeringsfond: https://gi.dk/" - "Dansk Byggeskik: https://danskbyggeskik.dk/" human_validation: "Guidance needs to be understood in relation to the specific building and its use. We review the result with you and agree which assessments we or your advisers are responsible for. The search can prepare the work, but does not replace a necessary site visit." responsibility_line: "AI can prepare, compile and flag uncertainty. The architect validates the consequences." last_reviewed: null last_updated: 2026-10-21 privacy_url: "https://fremtidenstegnestue.dk/en/about#privacy" related_pages: - url: "https://fremtidenstegnestue.dk/en/about#contact" relation: access - url: "https://fremtidenstegnestue.dk/en/about#method" relation: method status: published --- # Find information for your project. Canonical URL: https://fremtidenstegnestue.dk/en/toolbox Markdown URL: https://fremtidenstegnestue.dk/en/toolbox.md Entity type: static Language: en Author: Fremtidens Tegnestue Published: 2026-10-21 Last updated: 2026-10-21 Privacy: https://fremtidenstegnestue.dk/en/about#privacy ## Short answer Information about buildings is spread across articles, guidance and reports. We use artificial intelligence (AI) to find and bring together relevant information from five Danish publishers. You receive an overview with links to the original texts and questions to investigate further. ## Who it is relevant for architecture studios, municipalities, foundations ## What FT can prepare - Search and read across the five publishers’ websites and PDF files. The tool can also query three of their chatbots, which give automatic answers to questions. - You can trace information back to the original texts. Searching your own documents or other reports requires separate development. ## What the architect validates - Guidance needs to be understood in relation to the specific building and its use. We review the result with you and agree which assessments we or your advisers are responsible for. The search can prepare the work, but does not replace a necessary site visit. ## An example: from school to library. A municipality is considering turning a former school into a library. It and its advisers need information about what the new use requires of the building. We can help them find a starting point. An imagined task. We agree the questions, result and professional review with you. - You ask the question: What should we investigate before the school becomes a library? You tell us about the building, its future users and what you need to clarify. - We bring it together: We search and read across the publishers. You receive relevant information, links to the original texts and a list of open questions. - You choose the next steps: We review the overview with you. It can inform site visits, conversations with users and further professional investigation. ## How to get started. Tell us about your task. You do not need to know about AI or have a finished plan. - Write to Thais: Briefly describe what you want to investigate and who will use the result. - We agree the task: We check whether the five sources fit your question. Before starting, we agree what you will receive, what it costs and how to test the result. - We search and review: We use the tool and review the overview with you. You do not need to install anything. Using it yourselves requires agreed access and setup; the search takes place in a connected AI program. ## The tool and our next steps. The search tool is built. We also plan guides and instructions to help you use it in your work. ## Search tool — Prototype built Search and read across the five publishers’ websites and PDF files. The tool can also query three of their chatbots, which give automatic answers to questions. You can trace information back to the original texts. Searching your own documents or other reports requires separate development. ## Guides — Planned Practical guides to setup, choosing sources and using the results in your work. Setup instructions are already available in the MCP section below. ## Instructions — Planned Instructions that help AI follow the same workflow when a task needs repeating. We will specify steps, source requirements and where a professional needs to make an assessment. ## Five publishers. One place to begin. The tool uses the publishers’ search functions and retrieves relevant original texts. The material mainly concerns Danish conditions and is primarily in Danish. These links open the publishers’ websites. The content belongs to the publishers. We develop connections to their services; these are not partnerships. - [Bevica](https://www.bevicafonden.dk/) - [Bolius](https://www.bolius.dk/) - [Realdania](https://realdania.dk/) - [Grundejernes Investeringsfond](https://gi.dk/) - [Dansk Byggeskik](https://danskbyggeskik.dk/) ## What is MCP? When you use the tool yourselves, you ask questions in an AI program. Our tool lets it search the five publishers and retrieve texts for its answer. An AI program that can use tools in this way is called an AI agent. The connection uses MCP, short for Model Context Protocol. It is a shared, open standard for how AI programs and tools communicate. Our tool is called an MCP server because it receives requests from the AI program and sends information back. In the prototype, the AI program can search, retrieve web and PDF text, and query the chatbots at Bevica, Grundejernes Investeringsfond and Dansk Byggeskik. Both chatbot answers and the program’s summary need checking against the original texts. MCP does not ensure that answers are correct or that all relevant information has been found. - [Read the official explanation of MCP](https://modelcontextprotocol.io/docs/getting-started/intro) ## Use the tool yourselves: setup and instructions This section is for you or the person helping with installation. You need an AI program that supports MCP, access to our program code and Docker. Docker is a program that can run the toolbox on your own computer. Access to the code is arranged with us. Arrange access and the choice of AI program with us before starting. Copying the text below does not give you access. Once you have access, copy the instructions into your AI program. They ask it to follow our current setup guide and agree changes with you before anything is installed. Start with a general question without names, addresses or other confidential information. Searches and questions are passed to the publishers’ services. Help me use the Fremtidens Tegnestue toolbox for a task involving buildings. First check whether ft_sources, ft_search, ft_ask and ft_read are available in your MCP client. If they are not, explain what is missing. The prototype requires access to the private repository https://github.com/FremtidensTegnestue/vaerktoejskasse and a local Docker installation. If I lack access, direct me to contact Fremtidens Tegnestue using the contact link below. Do not invent a public endpoint. Follow the current README and CONTAINER.md, and agree installation with me before changing my setup. Contact: https://fremtidenstegnestue.dk/en/about#contact. Once connected, call ft_sources and explain which sources and tools fit my task. Then ask which building or workflow I want to explore. Use generic or anonymised questions only: queries are sent to the publishers' services. The sources primarily cover Danish buildings and are mainly in Danish. Search relevant publishers, read original passages and prepare a short source overview with key points, links and page references where available. Separate source statements, chatbot answers and your own proposals. Flag gaps, conflicts and historical material. A generic source does not establish the condition of a particular building. An architect or relevant professional must assess consequences and next steps. ## Sources and basis - [Bevica](https://www.bevicafonden.dk/) - [Bolius](https://www.bolius.dk/) - [Realdania](https://realdania.dk/) - [Grundejernes Investeringsfond](https://gi.dk/) - [Dansk Byggeskik](https://danskbyggeskik.dk/) ## Uncertainty and boundaries The search tool is built as a prototype: an early version you can test with us by arrangement. The content belongs to the publishers. We develop connections to their services; these are not partnerships. AI can overlook or misinterpret information. Older texts, drawings and figures need separate checks. Searches are sent to the publishers’ services and must not include confidential project information. The tool depends on those services being available. ## Next step Tell us about your task. You do not need to know about AI or have a finished plan. ## Related pages - [Talk to Thais about a trial](https://fremtidenstegnestue.dk/en/about#contact): access - [How we work](https://fremtidenstegnestue.dk/en/about#method): method ## Citation guidance When citing this page, cite the canonical HTML URL (https://fremtidenstegnestue.dk/en/toolbox) as the public source and use this Markdown URL only as the agent-readable representation. Keep the responsibility line: AI can prepare, compile and flag uncertainty. The architect validates the consequences.