The draft you write from scratch takes an hour. You have your templates, so it takes thirty minutes? What if the template came back filled in one minute, checked against the case file and the law, with every claim citing the exact madde it stands on? That's what this is: an assistant that reads the case, knows which laws matter, and does the work you'd rather hand to someone else. Below is one complete run on a fictional wrongful-termination case. Read what it read, watch what it did, open the document it produced.
Emre Kaya, five years at a logistics company, terminated the same day the letter was handed to him, for "performance below expectations." His file: the employment contract, the termination letter and a payroll slip. All the documents used as input were uploaded as phone photos. No scanner, no clean PDFs; court files are messy, and the system reads them anyway, Turkish characters included.
Two prompts: "is this termination valid?" and "draft the reinstatement claim."
Real session, real system. Output is in Turkish because the lawyer asked in Turkish; it answers in whatever language you use.
Every document from the demo case, exactly as uploaded, and the drafted filing. The case is fictional; the reasoning is not.
Every claim cites its madde, and the citation is verified against the loaded statute text, not the model's memory.
It computed the hard deadline. It checked today's date, then calculated: mediation application due by 26.07.2026, nine days out. The procedural trap in 7036 m. 3, mediation as a condition for filing, miss it and the case is dismissed, is exactly what it caught, unprompted, again in the drafting step.
It tells you what it doesn't know. The 30-employee threshold in m. 18 does not appear in any of the three documents. Instead of assuming it, it said so and logged it as a task for the lawyer.
It doesn't invent. The draft leaves counsel details and the mediation record as marked placeholders rather than making them up. It advises; it does not improvise.
It works your way. Your templates, your structure, your style of argument. The drafts come out the way your filings look, not the way a chatbot writes.
You talk to Claude, the AI most firms already know. Underneath, it runs on the top tier of Claude models and on the technology the whole AI industry is racing to build right now, working on your cases today. A few steps ahead of what everyone else offers.
Getting a case in takes five minutes: create the case, link the laws it turns on, upload the documents. Photos, scans, PDFs, Word files, Turkish characters intact, text extracted automatically.
Step 1: create the case. One case file per matter. Everything the assistant sees is scoped to this case; nothing leaks between cases or between firms.
Step 2: link the relevant laws. The statutes this case turns on get linked to it. This focuses the assistant on the right law first, and it can still search the wider library when something outside the linked laws matters.
Step 3: upload the documents. Photos, scans, PDFs, Word files. Text is extracted automatically, Turkish characters intact.
From here the lawyer works in the plain Claude chat window, and the run above shows the rest.
Behind the conversation:
Each firm gets its own isolated system: own database, own documents, own login. Nothing is shared between firms.
Claude connects to it through a secured, authorized channel. The reasoning happens in Claude; the documents and the law never leave the firm's own system.
Findings, deadlines and open tasks are saved back to the case file automatically, and the next session picks up exactly where the last one ended. This enables multiple users to work on the same case at the same time.
The full consolidated texts of İş Kanunu (4857), Türk Borçlar Kanunu (6098) and İş Mahkemeleri Kanunu (7036): 797 maddeler, loaded madde by madde from the official texts, every one individually addressable. The assistant is required to look the law up before making any legal claim. And the corpus is not fixed. More statutes, new amendments as they pass, a different jurisdiction, a different language: the system uses what your practice needs. Evidence not in Turkish or English is not a problem either; documents in any language are understood and searchable, and you still work in yours.
Where your data lives and what gets stored is your call: your own server if you want one, and nothing retained by the AI provider if that's your requirement. Nothing is shared between firms, and nothing from your cases is used for any further training of AI models.
This demo was not built for this page. It is the same system that runs in production on Serbian law, carrying a full national corpus of 1,590 regulations and 88,000 articles, loaded here with Turkish law instead. Different corpus, different language, same system. That is the design.