In the built-in chat or through the Bridge to your desktop AI.
IRONSTICK offers you two ways of working on your case files with an AI: the built-in AI chat and the Bridge to your desktop AI. Either way, the AI reaches your data exclusively through IRONSTICK — with the same tools and behind the same checks.
1. AI Chat and Bridge — the Difference
Two ways to the AI
The AI chat is built into IRONSTICK: you type in the IRONSTICK window, and IRONSTICK sends the request to the AI service with your own API keys — names and numbers leave your computer only as neutral placeholders. The Bridge works the other way round: you type in the desktop AI app you already use — Claude, ChatGPT, Qwen or Kimi Desktop — and that app works live on your case file, under your own account with the provider.
AI chat (built-in)
Bridge (desktop AI)
Where you type
In the IRONSTICK chat window
In your desktop AI app
Cost
API cost per call
Subscription — considerably cheaper for large case files
Before the first external request of a chat session
Before the first data access after each program start
Best for
Everyday work: questions about the case file, letters and pleadings in dialogue — always at hand, on the tablet too
In-depth work on large case files: reading entire cases, hundreds of documents, several drafting rounds
2. The AI Chat in IRONSTICK
AI configuration
AI configuration. The card with all AI providers.
Groq, OpenRouter and Mistral — free AI providers for background tasks such as e-mail prioritisation, the Data Consistency Monitor and the daily briefing; each can be switched on and off individually.
Qwen, Kimi, ChatGPT and Claude — the AI providers for the built-in chat, each with your own API key. If several are stored, IRONSTICK uses them automatically in the order shown; if one is unavailable, the next one takes over.
Your AI, your contract. You choose the AI provider and store your own API key — Claude, ChatGPT, Kimi or Qwen. Billing is directly with your provider.
Free background AI. If you wish, Groq, OpenRouter and Mistral take over routine tasks such as assigning e-mails and the daily briefing — each provider can be switched on and off individually.
Red/Blue Team. If you wish, a second review pass critically checks every AI result.
Desktop AI through the Bridge. Claude, ChatGPT, Qwen and Kimi Desktop work directly on the case file under your own subscription.
The built-in chat
Local first, then external — with your consent. The chat is the tightly controlled way to the AI.
Your computer first. Simple requests — opening a case, calling up a client — are handled by a local layer without an AI model and without internet.
Your consent first. Before anything goes to an external AI service, the chat asks you.
Pseudonymised. Names, file numbers, other numbers and addresses leave your computer only as neutral placeholders; the answer is translated back on your computer.
Logged. Every call to an external AI service is recorded locally.
3. The Bridge — Your Desktop AI on the Case File
Bridge
Full depth for large case files. Your desktop AI reads the current state of the case file — chronicles, full texts, e-mails, deadlines, master data — through more than 200 purpose-built tools and works in 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 printing on your letterhead.
“Compare the three expert reports and build me a table of contradictions.”
Included in your subscription. Reading entire case files and drafting in several rounds runs on the subscription you already pay for the desktop AI app.
Switched on by you. The Bridge works only when you allow access; every active AI shows its icon in the menu bar.
Several AIs, one case — your advisory board
A genuine second opinion. Claude, ChatGPT, Qwen and Kimi are different models with different training — and they can work on the same case at the same time, with the same state of the case file.
Develop case theories — one AI builds, the other stress-tests.
Harden pleadings — write, review, rebut, refine.
You conduct in natural language: “Work this out … give it to the other one for review … bring me its assessment.”
4. Verified, Not Taken on Trust
Why an AI alone cannot handle large case files
More than an AI chat in the browser. Give a language model hundreds of pages at once, and details from the middle get lost, the conversation overflows, and every session starts from scratch.
Structure first.IRONSTICK has already organised the case file: documents classified, data indexed, annexes linked.
Targeted access. The AI retrieves exactly the passages, documents, e-mails and deadlines the question needs — fresh from the case file.
A memory. Insights are retained across sessions; the AI builds up knowledge from its work on the case files dynamically.
Guided, not hoped for
Fixed checks instead of good wishes. Fixed checks stand between the AI and your case file — the facts are checked by plain program logic.
Read first, then write. The AI writes only from what it has demonstrably read.
Statutes from your library first. Legal provisions are cited from your statute library; the AI goes to the internet only when the library does not contain the provision.
Check before delivery. A pleading leaves the system only after passing the check — as a submission-ready template for your review and signature.
Proposals instead of writing. The AI only proposes changes to master data and memory; only what you approve takes effect.
Secret documents stay invisible. Entries marked secret never reach any AI.
Facts and prose
Two review paths for two kinds of content.IRONSTICK draws a fundamental distinction between facts and prose.
Facts — checked deterministically.IRONSTICK checks evidence numbers, dates, parties, file numbers and legal citations against the case file and the statute library using fixed program logic.
Prose — checked by the Red/Blue Team. A second AI critically reviews the line of argument, structure and wording.
The Red/Blue Team itself works with IRONSTICK. It has access to the case file and checks its own recommendations against statutes and facts in the background.
5. Nucleu and Snake Gate
Nucleu and Snake
Nucleu — the binding direction of a case. The Nucleu (atom icon in the case) sets down the core line of argument: the guiding facts and evidence, what must be said and what must never be said. Every document the AI writes for the case is subordinate to it.
On two levels. For an entire matter — valid for every case within it — and for the individual case.
The AI drafts, you decide. Every change appears marked in red; only your approval saves it as a new version.
Required reading. Without having read the current Nucleu, the AI cannot write, check or deliver any pleading for the case.
Final round of the review chain. Every pleading is finally checked point by point against the Nucleu.
Snake — the register of the opposing side's contradictions. The Snake (snake icon in the case) records every statement of the opposing side that the evidence in the case file contradicts — with document, date, case and page, and with every document that contradicts it.
Every line checked against the case file. Evidence number, date, case and page must be correct; otherwise the entry is rejected.
Built up through research or in the middle of a conversation — as soon as a contradiction surfaces, without duplicate entries.
The Snake Gate. Before a pleading is finished, the AI asks whether the Snake round should run. On your yes, every Snake entry of the client is checked and settled as adopted, discarded or not relevant — the opposing side's contradictions flow systematically into your pleading as evidence-based arguments.
Your approval for every version.
The AI works, you decide. Setup and workflows are described in the Bridge manual.
Get in touch
Tobias Werren
Scan the code to save our contact card — phone, e-mail and website.
Contact us directly by WhatsApp or Telegram, or call us. We speak English, German and Romanian.