PATRON · by MateMatic Solutions
Case-file analysis, legal research and document work - with source control, an audit trail and human oversight. PATRON runs locally, on your own hardware. Open source.
Open source (AGPL-3.0) · Local-first · Learns with you · Audit trail
Not a lack of will. A lack of trust. PATRON is built around the three fears we hear most often in law firms - and around an answer to each.
ProblemThe model writes falsehoods in an expert's tone. You sign off on it, not the model.
PATRONEvery claim drawn from the file gets a footnote to a verbatim quote and a page number. What isn't in the document, it doesn't invent.
ProblemUploading a case folder to an external service raises a professional-secrecy question.
PATRONIt runs locally, on your computer. Sensitive matters need never leave the machine at all.
ProblemYou have an answer, but not where it came from or how to defend it.
PATRONIt shows the source of every claim and records an audit trail you can check without having to trust us.
This is the heart of PATRON. When it relies on a document, it attaches a footnote to the source. You click the quote and see the verbatim passage with its page number - before you build a filing on it. What isn't in the file, it doesn't invent.
Hover over a footnote to see what PATRON shows beneath each one
The court found the clause unfair1QuoteContract, p. 14Passage found in the file word for word. Illustrative example., and the doctrine concurs2ParaphrasePATRON's summaryFaithfully rendered, but not a quote. Verify before relying on it.. The claim that the entire contract is void3No basisNot in the filePATRON flags it and does not pass it on. has no basis in the file.
The passage exists in the document word for word. You click the footnote, see it with its page number, and paste it into your filing.
When PATRON summarises or draws on general knowledge, it sets that apart from the file's content. Check it before you cite.
What it can't find in the document, it doesn't invent. It says plainly that there is no basis for it in the file.
The mechanism supports verification; it does not replace the lawyer's judgment.
Large matters don't fit into cloud tools. You cut the folder into pieces, spread it across several notebooks and lose the context between them. PATRON takes the whole matter at once - thousands of documents in a single Project.
It runs locally, so the limit is your disk and memory, not an account quota. A six-hundred-page scan goes in whole, without trimming.
The stronger the hardware, the larger the matter. Scale grows with your machine.
From a question to the folder, through field extraction, to your choice of model. Each feature kills a specific daily pain in the firm.
ProblemA question to a six-hundred-page folder means hours of reading.
SolutionYou ask in plain language and get an answer with a quote and a page number.
ProblemCase documents scattered across drives, emails and scans.
SolutionOne Project is one matter. You import a folder in one move; PATRON keeps everything in context.
ProblemCopying dates, amounts and penalties from dozens of contracts into Excel.
SolutionPATRON extracts fields from many documents into one table. Each cell links back to its source; export to Excel.
ProblemA scan or phone photo is a dead image, not searchable text.
SolutionIt reads PDF, DOCX, DOC, scans and photos (OCR) - from a court letter to a snapshot of a document. It indexes the whole folder for search.
ProblemSensitive data shouldn't reach the model uncontrolled.
SolutionPersonal-ID numbers and names pass through pseudonymisation; documents pass a security scan before the model.
ProblemA single AI provider means lock-in and a question of where the data lands.
SolutionYou connect your own model: Claude, Gemini, a local Ollama or OpenRouter. Sensitive matters stay fully offline.
Every matter starts from zero, and knowledge leaks away into emails and people's heads. PATRON builds a local memory that grows with your practice - it remembers documents, matters and the links between them. Like a personal Obsidian, except it understands case files.
Underneath, three layers work: vector search (RAG) over your documents, a local SQLite database for matters and decisions, and a knowledge graph linking parties, matters and sources. The longer you work together, the more context it has at hand.
All local, on your own hardware. The graph grows under supervision - a human approves the links within it.
The same lease-agreement review comes back every week, and every time you click through it from scratch. You save the sequence of steps as a workflow and run it on each new matter.
Set up a first-pass contract review once, then one click on a new folder.
Starting from a blank page is the same fight over wording every time. PATRON ships with built-in skills - a document reviewer, a devil's advocate, an editor - always on, locally. You start from a tested one and adapt it to your practice.
Add more skills from the MateMatic Boutique catalogue, one file at a time.
Searching across six open tabs means copy-pasting and losing citations. Six MCP connectors (open, MIT-licensed) link PATRON to authoritative sources - it reaches into them while answering.
Common, Supreme and Constitutional Court rulings, plus the National Appeals Chamber.
Administrative courts (Supreme + 16 regional, CBOSA).
Statutes and acts (Journal of Laws + M.P., Sejm ELI).
Company register: relationships, governing bodies, shareholders.
EU law and Court of Justice case law (live SPARQL).
GDPR, AI Act, DORA, NIS2, eIDAS 2.0, CRA.
Compliance asks "how did this answer come about", and all you have is a chat thread. PATRON records every interaction with the model in a hash-chain, with a Merkle tree above it. An auditor verifies consistency offline, without trusting us and without access to the content.
You export an audit package as evidence of due diligence under AI Act art. 12 - who, when, with which model, on which source.
MateMatic's reading of the AI Act / GDPR, not a position of the Polish bar associations.
The difference is architectural, not cosmetic. Where the case files live decides what a firm can promise its client.
| PATRON | Typical cloud legal AI | |
|---|---|---|
| Where data lives | The firm's own hardware (local-first) | The vendor's cloud |
| Sensitive matters | Can run fully offline on a local model | Sent to the provider |
| Source control | Quote / paraphrase / no-basis, per claim, with page number | Varies; often unattributed |
| Audit trail | Hash-chain + Merkle (RFC 6962), verifiable offline | Vendor-controlled, where present |
| Licence | Open source (AGPL-3.0) | Proprietary |
| Jurisdiction | 6 Polish & EU legal-source connectors built in | General-purpose |
"Typical cloud legal AI" describes a common architecture, not any single named product. Compare against your chosen vendor's current terms.
Is PATRON a European alternative to Harvey?
PATRON is a European, local-first take on legal AI. Where cloud platforms such as Harvey, Legora and others process a firm's documents on their own servers, PATRON runs on the firm's own hardware - case files need never leave the building - and it shows the source behind every claim. Different architecture, same goal: legal AI a firm can actually defend.
Is PATRON GDPR-safe?
PATRON is built local-first, so personal data can stay inside the firm. Sensitive matters can run on a local model (Ollama) fully offline, and personal-ID numbers and names pass through pseudonymisation before any model sees them. This is MateMatic's reading of GDPR and the EU AI Act, not a position of the Polish bar associations.
What is local AI for law firms?
Local AI runs the model and your documents on your own hardware instead of a vendor's cloud. For a law firm that means case files don't leave the office, professional secrecy becomes a property of the architecture rather than a promise, and the work can run fully offline. PATRON is built this way.
Will the AI make up something I have to sign off on?
PATRON attaches to every claim drawn from the file a footnote to a verbatim quote and a page number. What isn't in the document, it doesn't invent - and where there is no basis in the file, it says so plainly.
PATRON is free and open. MateMatic handles deployment, configuration and training.