IRONSTICK

1. The structural problem of SaaS — and the IRONSTICK answer2. The hallucination firewall — unique, to our knowledge3. Why IRONSTICK — and not a cloud legal AI4. Why now — and why not just ChatGPT?5. Your morning, in ninety seconds6. Instant case summaries — ask, don't dig7. From case file to a submission-ready template — in minutes8. Detail gain — a file that remembers everything9. A document system built for court10. Your desktop AI, plugged straight into your file11. Everything local — by construction11b. Three layers of protection — on from the first start12. More than one desk? The office server13. Full scalability — the system grows with the firm14. Self-healing and near-zero maintenance15. When hardware dies — disaster recovery built in15b. No IT department? That is the point of the design16. What this means in your week17. How you get started18. What you getGDPRDownloadPricingGet in touch

BridgeGDPRDownloadPricing

IRONSTICK

The case-management system with built-in file intelligence — for lawyers who want their evenings back.

IRONSTICK is a complete law-practice system for Windows and Android tablet: clients, cases, evidence, deadlines, e-mail, document production — with an AI layer that actually knows your files. Cleanly pan-European — and deliberately built for the continental-European civil-law tradition, the world of codified statutes: available in eleven languages — English, German, French, Italian, Spanish, Romanian, Portuguese, Dutch, Turkish, Ukrainian and Russian — the system recognises and builds out legal jurisdictions automatically, dynamically expands its legal database and keeps it up to date, and adapts itself to the needs and particular ways of its user. Offline-first, and your data never leaves your own computer without your say-so.

Hours become minutesPleading drafts, case summaries and dossiers are generated from the file itself — your job shifts from assembling to reviewing.

Nothing gets overlookedEvery document, e-mail, photo and spoken word is searchable — including scans and phone-call transcripts. Deadlines and silent cases surface by themselves.

Everything is one question away“What came in from the opponent last month?” — ask in plain language, get an answer grounded in the actual documents, with clickable sources.

You stay in controlLocal data, your own storage path, consent before any external AI call, pseudonymization, audit logs — and a written GDPR disclosure of every data flow.

What law firms really need — and what IRONSTICK delivers

Who IRONSTICK is made for

IRONSTICK pays off wherever many documents, many parties and hard deadlines come together — and where client data must not leave the building. The firms that gain most are those that lose hours today searching, typing and assembling.

• Solo lawyers and small firms. No IT department, no budget for technicians: IRONSTICK runs on an ordinary Windows computer, maintains itself and takes over the work that would otherwise need a second pair of hands.
• Litigators with extensive case files. Whoever handles thousands of pages of evidence queries the file in natural language and gets substantiated answers with sources. Court dossiers are built in five guided steps, Bates-numbered and indexed.
• Commercial, insolvency and criminal lawyers. These proceedings involve many persons, companies and entanglements. The network of persons and companies across all case files makes connections visible that could never be tracked by hand — right down to OSINT profiles of every person and judge profiles with an evaluation of their decisions.

Part I — Sovereignty: why not a cloud

1. The structural problem of SaaS — and the IRONSTICK answer

With cloud practice software you hand your responsibility and your data to strangers — and hope their cloud never burns. IRONSTICK means maximum sovereignty: the system is constructed so that even a total failure of the server — or a hijacking of the infrastructure — cannot stop the practice.

SaaS solutions sell the promise of security — this architecture delivers native, structurally enforced security. Classic cloud and SaaS practice solutions carry a gigantic structural problem their marketing brochures prefer to conceal: they are highly centralized. IRONSTICK is democratic and absolutely decentralized.

• The central single point of failure. When a SaaS provider is hacked — a supply-chain attack, captured cloud credentials, a configuration error at the cloud host — thousands of law firms are hit at once. This architecture distributes the risk physically across the individual office locations and their own hardened local networks.
• The liability and data-protection trap. With SaaS, highly sensitive client data sits on someone else's servers — Microsoft Azure, AWS, the vendor's data centres. After a data leak, a US-CLOUD-Act access or a server outage, the firm stands in the rain, professionally and under data-protection law. With IRONSTICK the data remains 100 % local-first on the firm's own devices.
• Held hostage by outages. If the SaaS provider's data centre goes down — or the provider simply has a bad day — nobody in the firm can work. Here, in the very worst case, the administrator buys a ~300 € replacement unit — or the practice simply keeps working autonomously on the workstations.
• No silent data loss. A single defect in a cloud database can erase data irrecoverably. This system rehydrates itself, in an emergency, from the physically separated workstations — including the laptops coming back from vacation.

2. The hallucination firewall — unique, to our knowledge

The one justified fear about AI in legal work: models invent citations, misquote figures, and claim to have read files they merely skimmed. The industry answers with prompts and disclaimers — IRONSTICK answers with architecture. Your desktop AI works on your file through 170 purpose-built tools, and every path to an outgoing document leads through deterministic gates: pure logic, no AI checking AI.

• Read proof. The system records, line by line, what the AI actually read of your file. Unread ranges are held against it in every answer — and delivery stays locked while gaps exist. “I read everything” stops being a claim and becomes a measurement.
• Citation provenance. Every evidence reference must point to a real record the AI demonstrably fetched and fully read in this working session — citing from search snippets or from memory fails verification.
• Statutes only from a verified library. Cited articles are checked against the letter of the law stored in your own verified library; new sources enter it only through your consent dialog. An invented article number blocks delivery.
• Zero tolerance for numbers and names. Amounts must match the record to the comma, party names to the diacritic; personal and bank identifiers run through checksum validators — an impossible CNP or IBAN can never reach a court.
• Delivery only against attestation. A formal pleading is refused unless a fresh verification of exactly that file state came back green — every edit voids the attestation, and five failed rounds abort the loop and notify you. Every delivered document carries its verification stamp.
• No side doors. A letter declared “informal” is content-checked and rerouted through the formal gates; renaming a document to reset its version count is refused; statute knowledge from press articles or the model's memory fails verification — only the law's own text, demonstrably read from the library in this session, grounds a citation; and every answer about a case carries the case's true subject from your records, so context invented from another case refutes itself on the spot.

The IRONSTICK way

To our knowledge, no legal software in the world enforces this combination. Observability tools log after the fact, sandboxes fence off access — IRONSTICK proves the work, at the only interface the AI has.

If you would like to know more about how the internal AI chat and the bridge for Claude, ChatGPT, Kimi and Qwen work, and which possibilities the IRONSTICK way opens up, click here.

The classic way

To this day, AI assistants work without any connection to the database holding your cases and client data — the only way in is uploading files. Most cloud-based systems work this classic way too: at best they shuffle text files around in the background — invisible to the user. At best, then, the AI draws only on the information in exactly these uploaded files. With a little luck they are not too large and the AI takes everything into account; otherwise hallucinations are pre-programmed. The AI cannot access the case itself: it works blind and tries to get the best result for the user. That can lead to dangerous misjudgements and false statements.

Additional skills and plug-ins change nothing about this. In the worst case they even make it worse: their extensive instructions take up a considerable part of the context window before the AI has read the first document — and the fuller that window is, the more unevenly the AI takes into account what is in it.

3. Why IRONSTICK — and not a cloud legal AI

The file stays yours. The AI comes to your data — not your data to the AI. That single sentence separates IRONSTICK from every cloud legal-AI platform on the market. In 2026 the industry moved en masse to one model: your client files live on a vendor's cloud, and an AI rented from that same vendor works on them there. IRONSTICK turns that model around — the case file stays on your hardware, and your own AI subscriptions connect to it through a controlled, documented bridge.

Three arguments, in the order your clients would rank them:

• Data sovereignty as a professional-duty argument. The database lives on your own medium; the software vendor never sees client data. Pseudonymization, audit log and the complete GDPR data-flow declaration ship with the product — including a ready-to-paste model clause for your client privacy notice. For professional secrecy and GDPR accountability this is not a feature; it is the answer.
• Bring your own AI — and keep the cost structure. You work with your own Claude, ChatGPT, Qwen and/or Kimi accounts (typically €20–40 per month each where a subscription is required) instead of paying triple-digit per-seat monthly fees for a cloud legal-AI platform. Over three years, for a small firm, the difference is measured in five figures — while the models you use are the same frontier models the big platforms rent to you.
• Controlled sharpness. Over 170 tools give the AI live access to the real file — chronicles, full texts, deadlines. But every write path runs through a review gate: proposed changes appear as a redline (strikethrough for deletions, bold for additions) and nothing is applied until you accept. Every reference in a generated pleading is checked against the file. The weapon is sharp; the trigger stays with you.
• Unmatched precision and speed. The IRONSTICK bridge gives the AI direct access to prepared data: every document already OCR-processed, classified, dated, referenced and case-bound. The AI never wastes time (or makes mistakes) reconstructing what a file contains — it queries the exact entry, receives the exact full text, and cites the exact reference. Large content moves by reference through the bridge instead of being retyped by the model, byte-exact and in seconds. The result is a level of accuracy and, above all, processing speed that copy-paste chats and cloud uploads simply cannot reach.

4. Why now — and why not just ChatGPT?

More and more lawyers are already experimenting with a raw AI chat — pasting documents into ChatGPT and asking away. For a single short document it works. Against a real practice load it runs into two hard walls — and both are exactly what IRONSTICK is built to remove.

• The context window. An AI chat can only work with what fits into its context window at once — at best a few megabytes of text (see the comparison in Part III). A grown case archive is many times that. So a raw chat forces you to cherry-pick a few pages, silently drops the rest, and starts again from zero every session. IRONSTICK removes the ceiling: the assistant fetches exactly the passages, documents, e-mails and deadlines it needs — fresh from the live file — and carries its distilled insight forward in a persistent, client-bound memory across sessions.
• Organisation. A raw chat has no structure — no clients, cases, entities, institutions, deadlines, chronicle or cross-case links. Every session is an island: nothing is filed, nothing carries over, nothing is auditable. IRONSTICK is the structure: the AI works inside your organised file, cites the exact exhibit and case number, honours the deadlines, and leaves a clean trail behind it.

The gap is simple: chatting with an AI, versus defensible, repeatable case work that still stands up months later.

And why external AI models at all — instead of a model of our own? The reason is economic. Operating an own, genuinely usable LLM would require an investment of 400,000–500,000 € — for servers, cooling installations and infrastructure alone — plus running costs of 70,000–80,000 € per year for operation and maintenance. And that does not even include the cost of technicians. For a law firm that is simply not sensible — the investment is far too high. And if IRONSTICK offered a cloud AI service of its own instead, it would in fact be exactly the same thing as the existing services from Claude, ChatGPT, Qwen and Kimi — only that IRONSTICK could never offer it at their prices. By plain logic, that makes no sense. That is why IRONSTICK connects your own subscriptions to the strongest models in the world instead of renting you a second-rate one of its own.

Part II — The daily practice

5. Your morning, in ninety seconds

Open IRONSTICK and the day is already sorted. The start screen shows what is due and what went quiet; the Agenda merges everything that happened across all mandates — yesterday's filings, overnight e-mails, new court dates, photos, transcripts — into one chronological feed. You see the whole practice without opening a single case. And beyond that: the separate daily briefing gives you a short spoken overview of the day — while you comfortably fetch yourself a coffee.

• Deadlines with a traffic light. The bell in the menu bar turns orange when something comes due, red when it is critical — permanently visible, one click to the full list.
• E-mail already triaged. The monitor fetches your inbox in the background, assigns mails to cases (a known registry number is matched without any AI at all) and prioritises the rest. New post announces itself in the menu bar.
• Silent cases and flare-ups. Cases with no activity for 14 days are listed so nothing quietly dies; cases that suddenly heat up — new deadline, imminent hearing, official post — carry a 🔥 marker until they calm down.

6. Instant case summaries — ask, don't dig

The case assistant answers questions about your files the way a good associate would — except it has read everything, and it answers in seconds:

• “Summarize the state of play in 5678/000/2026.”
• “Which letters came from the opponent, and did we answer the last one?”
• “Who said what about the site inspection — any contradictions?”
• “Have I already filed this exhibit with the court?”

Every answer is grounded in real documents — numbered hits you can open with a word (“open 1”, “read 1 aloud”). Routine questions are answered by a local layer on your own machine, without any data leaving the device; heavier analysis escalates to external AI only with your consent. The assistant speaks your language — all eleven of them — and can be driven entirely by voice, hands-free, with offline speech recognition and read-aloud.

7. From case file to a submission-ready template — in minutes

This is the flagship. You write pleadings and letters in dialogue with the AI — in IRONSTICK's own chat (on your API key) or through the desktop-AI bridge (Claude, ChatGPT, Qwen or Kimi Desktop). Compose text › With AI help in a case opens the chat with the case and the document already named; from there the AI works like an associate on the live file: client master data, the case chronicle with document full texts, deadlines, related exhibits, the annex list, your case instructions — and even the recent history of the client's neighbouring cases, because the opponent's answer often sits in the parallel file. It asks back when something is open instead of pressing your case into fixed masks.

• Research instead of assertion. While drafting, the AI actively searches the file. It is not allowed to claim “no reply was received” unless a search has proven it.
• Verification before hand-over. A deterministic scanner checks the finished draft against the live file: every evidence reference, every party, institution and file number, every statute citation against the verified law library. Contradictions block the hand-over; internal file stamps never leak into the document.
• Red/Blue Team, built in. On request a second model reads the draft as opposing counsel before hand-over: substantial findings send it back for rework, up to two rounds — and every point is analysed on the merits, never adopted blindly.
• A submission-ready template — never a finished filing. The result arrives through IRONSTICK's export dialog and lands in the built-in word processor with one click: A4 pagination, your letterhead (first page or every page), page-break control, signature placement, annex list taken over automatically — print, PDF, DOCX or Markdown. You review, correct and sign; the responsibility for what is filed stays with you.

Hours of file study and drafting become minutes of review — you read, sharpen, sign.

8. Detail gain — a file that remembers everything

The heart of IRONSTICK is evidence intelligence. Everything you capture — scanned PDFs, e-mails with attachments, WhatsApp histories, photos, audio recordings — becomes searchable full text. OCR is built in (English, German, French, Italian, Spanish, Romanian, Portuguese, Dutch, Turkish, Ukrainian and Russian — nothing to install) and the search is deliberately forgiving: diacritics, typical scan errors and partial names still find their target.

• Annexes and filings, proven from the paper itself. When a document is captured, IRONSTICK reads which of your exhibits it physically contains and which registry numbers it carries. The chronicle then shows, on every exhibit: filed with the court on … in case … — and on every outgoing letter, its full annex list. “Did I already file this?” has a one-click answer.
• Letter-number chains. Every registry number links all documents of the same procedure into a chronological chain — the documented history of who wrote what, when, and whether it was ever answered.
• Media with substance. Photos keep their capture location (one click to the satellite view), phone recordings become speaker-labelled transcripts — all of it searchable, all of it citable evidence.
• Cross-case awareness. Conflict-of-interest check on intake, warnings when the same opponent appears in several mandates, a relationship graph of all parties, and a relevance radar that finds related material you did not think to look for.

The network builds itself. A desktop AI on the bridge reads the file, researches the parties on the internet and files every person, company, institution and connection it finds as a one-click-to-approve proposal; a background auto-linker ties every party to every case whose documents name it; and the Universe turns the whole inventory into a fly-through 3D network with automatically detected clusters — dense, tightly interwoven groups pulse red. In hours you hold a highly condensed person-and-company network across your entire data set — impossible by hand — and the visualization points straight at the cluster.

What this means in practice — two days, measured. Over two working days in September 2026 a single desktop AI on the bridge read the files of a multi-party proceeding, researched the parties on the internet and filed what it found; the lawyer did nothing but read and click. The result, counted afterwards in the database: 293 persons and companies, 11 institutions, 811 relationships and 733 party-case links — 2,487 review decisions in total, 97 per cent accepted, 3 per cent rejected at the gate. By hand — sources read, every person identified, every connection checked, every record typed — that is about 310 working hours: eight full-time weeks for one person, or half a year alongside live mandates. And the hours are not even the point. No one holds 300 names, 800 connections and a dozen friend lists in their head; the clusters that glow crimson in the Universe are invisible in the filing cabinet. The 3 per cent are why the review gate exists: each one would have been a wrong record had it been written straight through.

And the same machine draws a client's case-net: cases as hexagons, grouped by your own matters, connected where they share parties, cite each other as an exhibit or name the other's case number — with four AI tools that read exactly this picture.

OSINT at the push of a button — built-in research on persons and companies

OSINT (open-source intelligence) — the systematic evaluation of freely accessible sources — is standard practice in investigative authorities and commercial law firms, but it is manual work: everyone researches differently, and everyone forgets something different. IRONSTICK turns it into a fixed procedure. The AI hands over name, city, district, country and whether it is a natural or a juristic person — the application does the rest itself, the same way every time, without gaps:

• Your own files first. A deterministic full-text search for the name across the entire inventory, all clients: per case the ten newest finds with paragraph and character position — and a note how many more there are.
• Case-specific research material such as commercial register extracts, land register extracts and the like: simply file it in the case file and tell the LLM to take it into account as well.
• Then the internet, on a fixed grid. CV, Facebook, LinkedIn, Instagram, city, district, scandals, press reports, declarations of assets and of interests, taxes, company holdings — for companies the commercial register and insolvency. The search terms are automatically in the language of the target's country: for Romania “declarație de avere”, not “declaration of assets”. This happens in the background via the client app, fully automated: the LLM receives a complete OSINT profile to work with. That is far more accurate and far cheaper.
• Three search engines at once. Google, DuckDuckGo and Qwant work through the grid in parallel; the hits arrive as one cleaned list without duplicates. If one engine blocks, the others keep delivering.
• A prepared profile — not a verdict. The AI receives both lists together with the search terms already covered, opens every find and every link one by one, sorts out namesakes and places only what is substantiated as a proposal in the data monitor. What is adopted is what you release.

The result in one to two minutes: a research profile that would cost a staff member half a day — and that is just as complete for the three-hundredth person as for the first. Only publicly accessible sources are evaluated.

This is what no one else has

We know of no other product that keeps a full relationship database of persons, companies and institutions at the level of the individual person, breaks it down to the single case file and the single exhibit, draws it cleanly for the lawyer and lays the same picture before the AI as text — together with a person analysis of 24 fields per person that can only be changed with proof.

Classic practice software keeps parties as an address book with a role in the case: no graph between them, no person analysis, no clusters, and their AI assistants read documents, not the network. The AI legal assistants on the market sit over the file in the provider's cloud, with a model you cannot choose. The large investigation platforms have the graph and the profiles, but as authority and corporate tools with a consulting project, without a case file in the lawyer's sense and without an interchangeable AI that reads them. A firm running criminal and civil proceedings with dozens of parties and interwoven cases falls between these markets. That is exactly where IRONSTICK sits.

Judges and actors — who with whom, and why

This knowledge does not apply to a single case only. It covers every party, every relationship and every role a person has ever played in one of your files, across cases and across clients — who sits on which board, who signs for whom, who appeared in which proceeding as witness, opponent or office-holder. It grows by itself, every day, while lawyer and AI work the files, and falls out as a by-product of daily case work.

• Relationship analysis — who with whom, and why. Every connection between persons, companies and institutions stands with the evidence that carries it; dense webs are automatically detected as clusters. You see not only that two parties are connected, but through what.
• Judge profiles. Every decision in your files is attributed to the judge or prosecutor who made it — from the heading and the signature block, not from guesswork. Added to this is the OSINT research on the person: career, public statements, connections to parties and law firms.
• Judgment evaluation per judge. How does a particular judge decide, what do they respond to, what do they take up from the pleadings and what do they pass over? IRONSTICK breaks every decision down into its parts and measures which share of the reasoning comes from which party's submissions — the basis for assessing a judge's line and predictability.
• Pleadings tailored to the judge. A pleading can be tuned to the individual judge or the panel that will read it — composition, prior decisions and positions, connections to the parties and, where present, grounds for recusal — because you know what runs in the background and who is with whom.
• Making cooperation visible. Any two pleadings, even from different proceedings and mandates, can be set side by side word for word: identical text blocks show which lawyers work together.

In court that is the difference between presenting a case and knowing the field: every document can be aligned to the actual web around the case — who is connected to whom, who is beholden to whom, who previously sat on which side — because the lawyer always knows where everything is. In this concentrated form this knowledge exists nowhere else; by hand it simply cannot be compiled.

9. A document system built for court

IRONSTICK is also a full DMS — one that thinks in exhibits, stamps and page numbers rather than folders:

• Capture with AI pre-fill. Drop a scanned PDF into the inbox: OCR runs, the AI proposes title, type, parties, case assignment, date and deadline — you confirm instead of typing. Every captured file gets a cryptographic integrity fingerprint, so later manipulation is provable.
• Stamping & pagination. Merging, Bates numbering, file stamps, metadata cleaning — every produced PDF carries a searchable text layer, so the finished document can be searched and quoted without ever re-running OCR.
• The dossier assistant. A guided, five-step path from chronicle to a submission-ready case dossier: choose entries, cover sheet and jurisdiction, preview with page thumbnails, then a finished PDF with table of contents and Bates numbers — plus an EDRM evidence ZIP for eDiscovery hand-over.
• Redaction that holds. Category switches (clients, names, places, amounts, ID numbers, IBANs …), a local finder that marks candidates, manual marking on the page — and on export the redactions are burned in irreversibly, with a clean searchable text layer over the top.
• Mail dispatch for oversized files. A large PDF is split losslessly into court-mailbox-sized parts, each with a ready-made delivery e-mail in the document's language — copy, attach, send.

Large-scale filings — 3,000 pages, stamped and paginated, in under 20 minutes

In civil procedure and large-scale litigation this goes straight to the biggest pain points of lawyers and their staff — time, error rate and manual labour:

• Massive time savings. What normally is a half-day to full-day project for a paralegal or the lawyer himself — assembling and merging PDFs by hand, checking pages, stamping every page manually (Bates numbering) — runs here fully automatically in about 20 minutes.
• Freedom from errors & consistency. With more than 2,000 pages, manual work quickly produces mistakes: slipped stamps, wrong page numbers on annexes, missing links, missed formatting rules. An automated pipeline guarantees an absolutely clean, uniform layout.
• No system crashes. Standard office software and ordinary PDF viewers go to their knees on a 700 MB package with 3,000 pages the moment you try to work through it fluently or re-paginate it.
• Competitive edge through speed. Under enormous deadline pressure — shortly before a claim or reply deadline expires — such a solution puts massive pleadings including all evidence on the table as a submission-ready template in record time.

Part III — AI at full depth

10. Your desktop AI, plugged straight into your file

For deep analytical work, IRONSTICK offers something no export folder can: a live bridge between your own desktop AI — Claude Desktop, ChatGPT Desktop, Qwen Desktop and/or Kimi Desktop, each a separately licensed add-on — and the running application. The AI reads the current file state through 170 purpose-built tools — chronicles, full texts, e-mails, deadlines, master data — and works at full depth:

• “Read the whole case and tell me where our position is weakest.”
• “Draft the appeal in our pleading format and hand it over to the word processor.” — the draft appears in IRONSTICK, ready for letterhead printing.
• “Compare the three expert opinions and build me a contradiction table.”

Built for large files — why a live bridge beats any file export. The subscribable bridge add-ons exist for one purpose: working on large data pools that demand analysis at full detail. They are highly controllable by you — switches, consent gates, per-AI licensing — yet every dialog and every evaluation stands on the entire IRONSTICK data pool as its foundation. That detail density is simply not reachable through a conventional file feed (exporting Markdown files into an AI chat): even the largest AI context windows hold only a few megabytes of text at once —

— while a grown case archive is many times that. The bridge sidesteps this ceiling: instead of stuffing files into the window, the AI is handed more than 170 dedicated tools with strict instructions on how to use them, and fetches precisely the passages, documents, e-mails and deadlines it needs, exactly when it needs them — fresh from the live system, across the whole pool. The context window then carries the analysis, not the ballast.

And when the task really is “read everything”: with a single call the AI stages the entire case — untruncated, OCR-normalized — into a local work area and then reads it piece by piece, without ever pushing the whole file through the window. A 400-document case becomes fully readable; truncation caps protect the chat, never the staged copy. The same area gives the AI a one-call session start — date, language, your current screen, the daily briefing, deadlines and open tasks in one protocol — and a freshness guard voids any staged extract the moment the underlying file changes: the AI can never argue from yesterday's data.

Why a raw LLM alone fails on large files — and why IRONSTICK flips the game

A language model on its own is simply outmatched by a grown case file — and a bigger context window (1 or 2 million tokens) does not fix that. This is the fundamental misunderstanding of the current AI wave:

• The “lost in the middle” phenomenon. Feed a model hundreds of raw pages at once and its hit rate for details collapses: facts from the middle of the file get “forgotten” — or hallucinated.
• The context-gagging effect. A single tool return flushing 50,000 characters of full text into the chat clogs the conversation: the model loses orientation, the context overflows, everything gets slow and expensive.
• No memory, no structure. A bare LLM knows nothing of deadlines, file numbers, parties, Bates stamps or the chronology of proceedings. It sees one long string of characters.

IRONSTICK does not make the LLM a pack mule carrying mountains of paper — it makes it a precision surgeon:

1. Structure first. Before the AI sees a single word, IRONSTICK has already structured the file: documents classified, dates indexed, OCR repaired, annexes linked.
2. Decoupling through tools & scratch. Via more than 170 specialised tools the AI fetches precisely what the question needs — and when a return value is huge after all, a hard limit automatically diverts everything beyond the connector limit into the local scratch vault — AES-256 encrypted on disk, readable only through IRONSTICK's gate. The chat stays perfectly clean.
3. Persistence & recursion. Through the shared AI memory and the automatic daily protocols no analysis result is ever lost: the AI builds knowledge over weeks instead of starting from zero with every prompt.

The bottom line: a raw LLM without a system is a brilliant jurist handed 5,000 loose, unsorted sheets and told to find the decisive line in five seconds. IRONSTICK is the infrastructure that turns that jurist into a functioning high-performance practice.

Set-up is one button per AI in the configuration; a skill plugin teaches the AI your binding output formats. What the AI learns about a mandate is kept in a shared AI memory — compact, client-bound, visible and deletable by you alone. And you always see what is happening: whenever an AI reads your file, its symbol lights up in the menu bar.

Several minds, one case — a built-in board of advisers

This is where it becomes more than a chatbot. Because the bridges can run at once, you put independent AIs on the same file and let them check each other's work — with IRONSTICK in the middle as shared database and coordinator. The decisive point: Claude, ChatGPT, Qwen and Kimi are different models with different training, so you get genuinely different opinions and approaches — a real second opinion, not one model nodding at itself. Today these are among the strongest AI models in the world; IRONSTICK puts them, side by side, on your case.

You direct them in plain language, like a partner steering two associates: „work this out… now give it to the other one to check… bring me its assessment… revise." The draft, theory or pleading is handed over through IRONSTICK, the second AI reviews it critically — tasked to find weaknesses, counter-arguments and errors, not to praise — and its assessment comes back to you. Both stand on the identical case data, and the entire exchange is logged (forensically clean).

This two-headed adviser/forensic team is ideal for:

• Developing theories of the case — build one, have the other stress-test it.
• Pattern recognition across the file — two models catch what one misses.
• Working out tactics and strategy — proposal and counter-proposal.
• Drafting and hardening pleadings — write, review, rebut, refine.
• Cross-checking your general lines of reasoning — before the opponent does.

A file that heals itself — you as the precision instrument

Old capture mistakes and botched scans stop being permanent. When the AI hits a stored full text that is truncated or ruined by a bad OCR run, it does not quietly work around it — it heals the record: it requests a fresh, page-accurate re-scan of the original PDF (you pick the pages in the familiar page picker), compares the new scan with the stored text passage by passage, repairs only mechanical scan damage — never rephrasing; numbers, names and amounts stay untouched — and hands the healed full text back. You see it in full, and only your click saves it. The archive gets cleaner with every week you work in it.

Where even a fresh scan hits physical limits, the AI turns you into its precision tool — Human-in-the-Loop Micro-Sourcing: instead of a useless “file could not be processed”, it opens the entry, names the exact spot (“page 12, near the top”) and asks you to type off precisely the three illegible words — or, within hard limits (10 MB, 90 pages), to hand it the original PDF via the paperclip so it can read the scan with its own vision. You never comb through a 200-page file; you are called in for exactly the 0.1 % a machine cannot resolve with certainty. Traditional systems know two states — “success” (however wrong) and “error, please process manually”; this is a symbiosis of AI speed and human eyesight that, in this form, exists nowhere else on the market.

Part IV — Data security: your data, your rules

11. Everything local — by construction

• Offline-first. The entire file lives locally, in a directory you choose — an encrypted USB drive, if you prefer. No forced cloud, no vendor server, no account requirement. Without internet, the file, search, documents and local assistant keep working.
• AES-256 encrypted database. The file that holds your entire practice — clients, cases, chronicle, e-mails, deadlines, the AIs' memory — is encrypted on disk with AES-256, every page individually and integrity-sealed. Nothing to set up, no password to remember: the clear text exists only in memory while IRONSTICK runs. Whoever copies the file from a disk, a stick or a backup holds ciphertext.
• AES-256 encrypted documents. The document store itself — every original PDF, scan, image, recording — rests encrypted as well, each file with its own integrity seal. Clear text exists only for the documents you are actually working on, in the temporary folder, and is wiped when you close IRONSTICK. What you export or send leaves the vault deliberately.
• Authenticator access code. Every start asks for the six-digit code from your phone's authenticator app — the same one-time codes your bank uses. No password to remember, nothing stored: the secret is derived from your license key at run time, and the license key gets you back in if the phone is lost. Until the code is entered nothing is shown, and the AI connectors receive no case data. The configuration file with your API and mail credentials rests AES-256 encrypted as well.
• Pseudonymization before external AI. When in-app functions do use external AI, names, ID numbers, case numbers, IBANs, phone numbers, e-mail and addresses from your records are replaced by neutral placeholders on your machine first — the answer is translated back locally.
• Consent and audit. External AI runs on explicit consent and is logged — the audit trail stores exactly what actually left the machine.
• Secret documents (Windows). A hidden, password-encrypted compartment for documents only the lawyer may ever see: invisible in every list, excluded from every AI channel and every export, never synced to the tablet — with an emergency destruction sequence for the worst day.
• Backups and sync without cloud. One-click full backup to wherever you want; Windows ↔ tablet sync runs over a USB cable, checksum-verified. Your phone receives calendar entries only — never file contents.
• Written disclosure. A complete GDPR data-flow declaration ships with the system — every external transmission, its exact content, and where the boundaries lie. Including a model clause for your own client privacy notice, in all eleven languages — English, German, French, Italian, Spanish, Romanian, Portuguese, Dutch, Turkish, Ukrainian and Russian.

11b. Three layers of protection — on from the first start

All protection layers from Chapter 11 are active out of the box — with nothing to configure and no switch that can go off by mistake: data at rest AES-256 encrypted — database, every document, the configuration and the AI's entire workspace; in plain words: no case data lies unencrypted on your disk any more — not even the AI's working files. Access at run time only with the code from the authenticator app. Data in motion only as your deliberate act — every path documented in the GDPR declaration that ships with the product.

Cloud legal software sells you the promise that someone else keeps your data safe. Here the protection is part of the construction, on your own hardware, and you can verify each layer yourself.

Part V — Growing with the firm

12. More than one desk? The office server

For practices with several lawyers, IRONSTICK offers the IRONSTICK SERVER — a separately licensed add-on that turns independent workstations into a coordinated multi-user practice, without a cloud and without giving up the local file. It runs as a native package on a Synology NAS — the small, inexpensive, near-zero-maintenance office boxes that already sit in thousands of law firms — on any current model, entirely inside your own network. The server never contacts the internet, never contacts the vendor, and carries no AI traffic.

Why a Synology NAS — and why that is not an IT project. We chose Synology deliberately because these boxes are plug and play: connect power and network, and they simply work. Installing the IRONSTICK server package on it is done in three minutes and then it just runs — no console, no configuration files, no administrator. The workstations are just as undemanding: the installer is a completely guided setup without IT knowledge and without lengthy questionnaires — start it, enter the license key, done. Clients and server find each other in the network automatically; nobody types an address. And a single-desk practice needs no NAS at all — the server is an optional add-on for firms with several lawyers.

• Share cases, keep your file. Release a case to a colleague with one action; they subscribe and receive a complete copy into their own local file. Revocable at any time; documents marked secret never leave your desk. And the real force multiplier: several lawyers — each with their own AI — can work the same case simultaneously, sending the owner messages and finished analyses straight out of their LLM. In an emergency, an entire firm can converge on one case.
• One legal library for the whole firm. Every statute text a colleague verifies and releases becomes available at every desk — and to every desk's AI: the assistant and the desktop-AI bridges cite the same verified, verbatim text from the shared library instead of fetching the web again or trusting model memory. Review dates, red flags for stale entries and automatic reassignment when a contributor is absent keep the library alive — one colleague's research becomes the whole firm's verified knowledge. A democratic system keeps it honest: every workstation re-checks its own contributed sources after two weeks — as an automatic background task, without any action by the user. If a contributor is absent at check time, the server distributes their checks among the other workstations; after four weeks a source is coloured red. Once in the system, it stays in.
• A firm-wide time overview. Every workstation reports its measured working time in the background; the server's admin pages show time per case, per client and per day — for a single user or the whole practice — plus presence, case-sharing history and a security log.
• Administered in a browser. Registration approval, licence, users, laws and time — all on one admin page served by the NAS itself. Yearly licence with a defined number of seats.

Case hand-over and subscription — how it actually works

Releasing a case is a deliberate act, never a side effect. The owner shares the case from within the app; it travels to the server in a two-phase transfer — metadata first, file content second, so an interrupted transfer never leaves a half-case behind — and large files are content-addressed, so identical documents travel and are stored only once. The server assigns fresh internal IDs to everything it stores: a received case can never collide with data already on a colleague's workstation. Colleagues see the shared case on their own server-cases screen and subscribe; the subscriber receives a complete copy into their own local file and works on it with the full system — search, assistant, AI drafting, everything. The owner can revoke the share at any time, the administrator can end any subscription centrally, and every share, subscription and export is recorded with date, time and address. Documents marked secret are excluded by design — they never reach the server.

The mega advantage: simultaneous AI-assisted teamwork on one case. Every subscriber works their copy with the full system — including their own desktop AI. Several lawyers can therefore attack the same case at the same time, each with their own LLM reading the file at full depth — and send the case owner messages and finished evaluations directly out of the AI session: a contradiction table from one desk, a hardened draft from the next, a risk assessment from the third, all flowing back to the owner while the work happens. When a deadline suddenly looms or a hearing is moved forward, an entire firm can converge on a single case overnight — parallel analyses, divided workstreams, one owner keeping the file.

A legal database that grows by itself — wired straight into the AIs

Every IRONSTICK desk maintains verified statute texts as full-text Markdown, each entry tagged with jurisdiction, source link and review date. The assistant and the desktop-AI bridges reach this library through three dedicated tools: a fuzzy search (the jurisdiction is a mandatory input — legal orders are never mixed), the verbatim retrieval of the stored text (the AI cites the verified wording, not its training memory) and the submission of new sources (whenever the AI encounters a relevant legal source online, filing it is a duty, not an option — link only; the application downloads, converts and verifies the text itself, and you release it through the data-monitor review gate).

This is what makes the database dynamic: it grows out of the firm's daily work. Every research session of every colleague — human or AI-assisted — feeds it. With the office server, each released statute is de-duplicated centrally (same source updates the entry; a newer version of the same text wins, the superseded one is shelved, not lost) and mirrored to every desk. Review dates keep it honest: entries unchecked for over a month turn red on every desk and in the web admin, and if a contributor is absent, the server reassigns their entries evenly among the active users — the review duty never dies with a vacation. The result: all the firm's AIs argue from one verified, living statute library — verified once by one colleague, cited verbatim at every desk, maintained under a common duty — instead of every desk fetching the open web again and trusting whatever came back. And it is civil-law thinking end to end: the authority is the codified statute text, cited verbatim — precisely how continental-European courts expect to be argued to.

Central mail intake — the Central Scan App

Available only in connection with the office server: the Central Scan App, a stand-alone program for the office clerk's PC — no IRONSTICK installation needed there. Up to three Central Scan licences can be acquired per server — each licensed separately — so up to three scan workstations can run per office; all of them may even work the same intake folder in parallel without colliding (a document opened by one clerk is automatically locked and greyed out for the others).

• One scan point for the whole firm. The clerk scans the entire morning mail into one folder and works the list document by document.
• Searchable by design. Pages that already carry a text layer are left untouched; only scanned pages run through local OCR, and the recognized text is embedded into the PDF as an invisible layer — the receiving lawyer never has to OCR it again.
• The recipient is recognized — without any AI. The app matches the document text against the case numbers and lawyer names of the firm (reported live by the workstations) — plain deterministic matching, zero tokens, nothing leaves your network. Recognized recipients are suggested; one click sends.
• Delivery is invisible. Each lawyer's IRONSTICK collects its own documents from the server automatically and drops them into the normal document intake — the usual capture workflow takes over. The server keeps a document only until the recipient has confirmed receipt; local copies at the scan desk clean themselves up after three days.

Time tracking — local on every desk, aggregated for the firm

Locally, IRONSTICK measures working time automatically as a side effect of working: no stopwatch, no start/stop discipline. Time is derived from actual activity in the application, attributed to the case you are working in, with idle time capped so an open window over lunch does not bill anyone. Each desk has its own time page — today, 7, 30, 90 days or everything; per case, per client, day by day with bars.

With the server, every workstation reports these measurements in the background — every five minutes, as compact day totals per case (including the client name), with missing days back-filled automatically; a desk that joins later delivers its complete history, so the firm-wide picture is whole from day one. The web admin's Time page is a faithful copy of the desktop time page, extended by one selector: a single user, or ALL — the whole practice. Time per case, per client and per day, across every desk — an objective basis for billing, capacity and management decisions, produced without anyone filling in a timesheet.

13. Full scalability — the system grows with the firm

The system grows with the firm. Started out as a solo lawyer, a hundred-strong team today — IRONSTICK grows along. What is needed? One IRONSTICK desktop client at the beginning as a solo lawyer, perhaps a bridge for Claude/GPT. Then two lawyers join forces: one more client and one more bridge are licensed, plus the IRONSTICK SERVER — up to five workstations are included. From then on, every additional colleague means one client and one bridge, and from the sixth workstation on, one additional server seat. No big investments into a distant future — licence only what you need right now. And if the team gets smaller, you simply buy fewer licences.

Part VI — Reliability: self-healing, backups, almost no IT

14. Self-healing and near-zero maintenance

Maintenance: the system is absolutely self-healing and therefore extremely low-maintenance. It runs absolutely stable and fluid even on hardware that is no longer quite up to date. And through the backup option, in an emergency you simply reinstall and play the backup back in.

The system repairs itself. The only dependency left is how fast the administrator walks to the shop for a replacement device, puts the software back on and plugs it into the LAN (server) — or plays the backup back in (workstation). Everything else takes care of itself. And even while the server is down — for whatever reason — the workstations simply keep working locally on their own computers: every lawyer has the complete file on their own machine, so nothing stops.

15. When hardware dies — disaster recovery built in

A server that coordinates a whole practice must be allowed to die without taking the practice with it — and this one is. The entire server state lives in one data folder on the NAS: include it in the NAS backup (e.g. Synology Hyper Backup) and a dead unit is restored in three steps — install the package on the replacement, put the folder back, start. Everything returns exactly as it was and the workstations reconnect on their own. And because the truth of the system is decentralized anyway — every workstation keeps its complete local file, its subscribed copies, its legal-library mirror and its own backups — even the worst case, a total NAS loss without any backup, heals itself: workstations re-register automatically, owners re-issue their case releases automatically (checked twice a day), subscribed copies are protected for 14 days instead of being discarded, the legal library regrows from the workstations' own mirrors, and pending case hand-overs cannot lose data by design. For even higher demands, the same package runs unchanged on the large, fully redundant Synology models — RAID disk redundancy, redundant power supplies, High-Availability pairs with automatic failover. Moving the office is just as undramatic: take the workstations with you, set up the NAS at the new address, connect everything to the LAN — Plug and Play, done. The workstations find their server again on their own; no fixed addresses, no reconfiguration, no IT appointment — and no expensive technicians, ever.

15b. No IT department? That is the point of the design

A fair objection: IRONSTICK runs on your own hardware, so there is no cloud vendor whose support desk fixes things when a local disk dies. True — and deliberate: nobody at a distance holds your files, and nobody at a distance can lose them, read them or lock you out. What that actually demands of a practice without an administrator is smaller than it sounds.

• One habit, nothing more: press Backup. One click on the backup screen writes a complete backup as a single ZIP; you save it wherever you like — a USB stick, an external disk, the office NAS. IRONSTICK does not leave this to discipline alone: if the last backup is older than five days, the daily briefing ends with an urgent reminder, spoken and written, with a direct link to the backup area; on the office server the administrator page shows every user's last backup date, red after five days.
• The backup itself is harmless to lose. Database, documents, configuration and the AI's working files are AES-256 encrypted — a backup ZIP on a lost stick is ciphertext, nothing else.
• Recovery is a routine, not a project. New or repaired machine: install from the stick, open the backup screen, play the ZIP back in, start. Everything returns exactly as it was — cases, documents, chronicles, AI memory, law library. No technician, no ticket, no waiting.
• Nothing to administer in between. No server to configure, no database to tune, no user accounts to manage — the license key and the authenticator code are all the credentials there are. The system repairs itself, workstations find the office server on their own, updates are the installer again. If you can save a document to a USB stick, you can run IRONSTICK safely.
• Second copies come for free where you use them. Every workstation in an office-server setup keeps its complete local file, the tablet sync carries a checksum-verified copy with you, and the NAS is backed up by its own tools (e.g. Synology Hyper Backup). Losing one device never means losing the practice.

The IT affinity required is that of any Windows workplace — not more. What IRONSTICK removes is the dependency on someone else's data centre; what it asks in return is a backup stick you actually use.

Part VII — In practice

16. What this means in your week

17. How you get started

• Modular licence. The base system (case file, documents, deadlines, e-mail monitor, word processor) plus separately licensed add-ons: the Claude Desktop bridge, the ChatGPT Desktop bridge, the Qwen Desktop bridge, the Kimi Desktop bridge, and the AI case assessment. All modules are unlocked in the same licence key — buy only what you use and add modules later with a licence key.
• Professional onboarding. Installation and structured training by a trainer follow a fixed curriculum — including the mandatory briefing on how to work with AI results responsibly. Training is licensed separately in three modules: Use of the IRONSTICK app, Use of the IRONSTICK app and office server, and Use of the IRONSTICK app and the LLM bridges. All modules are delivered on site: in one- and two-lawyer practices each module takes one day and is taught directly on the live case files — with the trainer under NDA. In larger firms, modules 1–3 are first taught in groups as a lecture with worked examples. Where needed, module 3 can be extended over several days to further raise the firm's efficiency in working with the LLMs. The bridges are built to be low-maintenance: a one-click recycle button restarts all connected desktop AIs and self-heals their plugin configuration.
• Gradual adoption — no big-bang migration. IRONSTICK slips into your existing routines instead of replacing them overnight. New cases are handled in IRONSTICK from day one; old cases are migrated only if and when you judge the effort worthwhile — case by case, at your own pace. There is no massive one-time migration project between your current way of working and the IRONSTICK way of managing a case file.
• Your scanner feeds IRONSTICK directly. Point your scanner or copier at one specific folder — nothing to migrate, nothing to install on the scanner side. When the mail arrives, you simply scan the whole stack in one go; every document appears in IRONSTICK's inbox of documents awaiting capture, and you then work through it comfortably, one document after another. Capture itself is AI-assisted: with the built-in OCR, registering a document takes on average no more than fifteen seconds.

18. What you get

• IRONSTICK for Windows — the full system; Android tablet app on the same file via cable sync; calendar reach-through to any phone.
• Eleven languages throughout — interface, assistant, and generated documents follow each case's jurisdiction.
• Five illustrated manuals (installation, daily operation, documents & e-mail, AI, writing with AI & word processor).
• The desktop-AI bridges (Claude Desktop, ChatGPT Desktop, Qwen Desktop and/or Kimi Desktop, separately licensed add-ons, parallel operation supported) with set-up button, skill plugin and their own guides — plus a locked-down Claude Code package for long analyses.
• The GDPR data-flow declaration with a ready-to-paste model clause for your client privacy notice.
• The office server (separately licensed add-on) for practices with several desks — case sharing, a firm-wide legal library and a firm-wide time overview on your own Synology NAS; see Part V.

System requirements

Optional: an Android tablet (Android 11 or newer, 64-bit, tablet-size screen) on the same file via cable sync, and your own AI account (for the in-app external AI and/or a desktop AI — Claude Desktop, ChatGPT Desktop, Qwen Desktop or Kimi Desktop) — without one, IRONSTICK simply runs as a fully local system.

System requirements — the office server (optional add-on)

IRONSTICK was not built to impress in a demo. It was built so that at 19:00 you close the lid — with the file read, the draft hardened, the deadline met, and the proof of all three in the system.

Get in touch

Interested in IRONSTICK, a demo, or the price list? Write to us.

Tobias Werren

Scan the code to save our contact card — phone, e-mail and website.

Feel free to contact directly via WhatsApp or Telegram — calls are welcome too. Spoken languages: German, English, Romanian.

IRONSTICK is legal-practice software. This website is for informational purposes only and does not constitute legal advice.

© 2026 IRONSTICK.