DOC.ai is our new product line for AI-driven tools across construction, energy and real estate. We pair large language models such as Claude and GPT with the data model we have shaped over 25 years of construction software - and with workflows that engineering and planning offices actually use. First concrete product: the Tender Agent for public construction tenders.
There are plenty of AI demos. Few of them deliver what site management, tendering and asset management actually need day to day: traceable results, clean data protection and a genuine grasp of how a tender procedure, a defect inspection or a construction file is really put together. DOC.ai grows out of exactly that experience.
The LLM is replaceable, the data model is not. From 25 years of construction software we know how tender documents, defect lists and construction diaries are structured - and how eligibility and award criteria need to be read. That industry knowledge sits in the workflows, not in the model.
Hosting in Germany, data processing agreement, clearly mapped data flows and Standard Contractual Clauses for US vendors. You always know which data reaches which provider - or you run the model on-prem and resolve the question internally for good.
Every AI output is a draft, never a final product. Reports link sources, scores are traceable, recommendations are reasoned. No black-box answers without evidence - the professional responsibility stays with your team.
Already running a Claude Code subscription or an OpenAI API account? That is exactly the licence we use. We do not retrain yet another AI of our own - we configure the data model, the prompts and the workflows. The AI core stays current, and you do not pay twice.
DOC.ai launches in 2026 with the Tender Agent. Three further solutions follow step by step: defect classification, construction-diary assist and reference matcher. Each solution sits on the same DOC.ai layer - data model, workflows, security and licensing logic are shared across the products. That is how the investment scales: anyone who starts with the Tender Agent can add the next solution without a new stack.
We do not build demos for the stage; we build tools for working days. The scenarios below describe what DOC.ai is being used for today or can be used for in the very near future - in real engineering offices, with real tender packages and real defect lists.
A new tender folder arrives: 40 PDFs, three form-sheet packages, an Excel with eligibility criteria. The Tender Agent classifies, extracts deadlines and requirements - and the bid lead can take a well-founded decision on whether to apply.
Required minimum turnover, certificates, reference count and staffing are matched against the company profile. Hard criteria via rules, soft criteria via AI scoring. The result: a clear eligibility score in per cent with a traffic-light recommendation.
Instead of starting from a blank page every time: the methodology generator produces an initial draft from your company profile and the requirements extracted from the tender. The subject-matter author then polishes the final text.
After the site walk, 240 photos come back - with the keywords captured in DOC.mobile. The AI proposes building element, trade and severity. Site management confirms or corrects each row. Prep time drops sharply.
The site manager dictates on the drive home: weather, crew, incidents, a note on a delay. Construction Diary Assist places the spoken text into the DOC.mobile fields - the office only corrects, no longer transcribes.
The Reference Matcher reads the tender requirements - then searches your reference database for the closest fits: type of work, volume, period, client type. No more recycling the same boilerplate reference block.
As-built documents, area schedules and equipment lists arrive in every format imaginable. DOC.ai structures the building data and prepares it for import into the technical due diligence - the assessment starts with complete master data instead of re-typing.
The maintenance tasks for technical installations are buried in operating and maintenance manuals spread across extensive documentation. DOC.ai analyses the documents, extracts the tasks and prepares them for import into maintenance planning.
Answers to the topics that come up most often in first conversations: models, data, licensing, GDPR, integration. If your question is not covered - drop us a line.
The default is Claude Sonnet (for speed) or Claude Opus (for harder analyses) - called through the Claude Code CLI against your existing subscription. On request we use the OpenAI API (GPT-4o, GPT-4.1) or locally hosted models such as Llama 3 or Mistral. The model is a configuration question, not an architecture question - DOC.ai workflows run on all three variants.
In standard mode every request reaches Anthropic (for Claude) or OpenAI (for GPT) for processing. No training data is produced from it - both vendors explicitly exclude this in their terms of service. If you do not want to take that risk, run DOC.ai on-prem: no request then leaves your data centre. We discuss with you which mode suits which kind of information.
For the cloud variant: yes. You need access to an AI service - we recommend a Claude subscription (the "Claude Code Subscription"), because for regular use it is considerably cheaper than paying per request through the OpenAI interface. You do not have to deal with the technical setup yourself: your IT department handles it once, and after that sign-in runs in the background. If you choose the on-prem variant - running everything in-house - you need no subscription at all, just a powerful computer with a graphics card (GPU) on which the AI model runs locally.
DOC.ai is a standalone product line - but we integrate it with DOC.mobile through the REST API. Defect Classification feeds results back into DOC.mobile; Construction Diary Assist writes into the mobile construction diary; over time DOC.ai will pick up tasks inside the workflows of DOC.project. You stay in one platform world.
We assume that it will - and we engineer the workflows accordingly. Every DOC.ai output is a draft, never a final product: reports link sources, GAP scores are traceable, recommendations are reasoned. Your subject-matter owners review, correct and sign off. Hallucinations therefore do not break trust - they show up as identifiable data points.
Yes. A data processing agreement with DOCby.net is standard. For US vendors (Anthropic, OpenAI) Standard Contractual Clauses apply. Data flows are documented - you always know which fields reach which provider. If you want to play it perfectly safe, run DOC.ai on-prem: then no processing happens outside your own data centre.
DOC.ai is licensed per user per year. The Tender Agent is designed as an entry product and is economic even at small team sizes. We provide concrete pricing on request - calibrated to the number of users, the model mode (cloud or on-prem) and the depth of integration. Get in touch: Contact.
Rarely necessary - and usually not the right route. We configure the data model, the prompts and the workflows. The general language and world knowledge comes from the LLM and stays current because we can switch to the latest model at any time. Where genuinely domain-specific terminology is needed, example data can be brought in as few-shot templates - without classic fine-tuning.
We will walk you through the Tender Agent using a real tender - your data, your workflows, your GDPR questions. No sales pressure. Response within 1 business day.
