PATRON is an open-source, local-first AI platform for law firms, built to support case-file analysis, document review, and legal research while keeping all data on the firm's own hardware. The platform is developed by MateMatic Solutions, a Kraków-based company specialising in AI architecture for law firms and compliance teams, and is released under the AGPL-3.0 licence.

Key features and functions include:

Source Control
Every claim that PATRON draws from an uploaded document is tagged with a footnote identifying whether the passage is a verbatim quote (with page number), a paraphrase, or a statement for which no basis was found in the file. Users can click any footnote to view the underlying passage before relying on it in a filing.

Local-First Architecture
PATRON runs on the firm's own hardware rather than a vendor's cloud. Sensitive matters can be processed entirely offline using a locally hosted model, with no requirement for data to leave the firm's premises.

Bring-Your-Own-Model Support
Users can connect their own AI model, including Claude, Gemini, OpenRouter-hosted models, or a locally deployed Ollama instance. This allows firms to select a model appropriate to the sensitivity of each matter.

Projects and Scale
Case files are organised into Projects, each representing a single matter. PATRON processes thousands of documents within a single Project, with no imposed file-size limit; the practical constraint is the firm's own hardware.

Document Import and OCR
The platform ingests PDF, DOCX, and DOC files as well as scanned documents and photographs, using OCR to make image-based content searchable. The entire folder is indexed for retrieval on import.

Table View and Field Extraction
PATRON extracts structured fields — such as dates, amounts, and penalty clauses — from multiple documents simultaneously and presents them in a single table. Each cell links back to its source document, and the table can be exported to Excel.

Pseudonymisation and Input Scanning
Before content reaches the AI model, personal identifiers such as national ID numbers and names are pseudonymised, and documents pass a security scan.

Audit Trail
Every model interaction is recorded in a hash-chain with a Merkle tree (RFC 6962) above it. The resulting audit package can be exported and verified offline, independently of MateMatic, and is designed to support due-diligence documentation under Article 12 of the EU AI Act. MateMatic notes this reflects its own reading of the AI Act and GDPR, and not a formal position of the Polish bar associations.

Knowledge Graph and Memory
PATRON builds a local memory layer across matters using three components: vector search (RAG) over the firm's documents, a local SQLite database recording matters and decisions, and a knowledge graph linking parties, matters, and sources. Links within the knowledge graph require human approval before being added.

Workflows and Skills Library
Repeatable task sequences — such as a standard first-pass contract review — can be saved as workflows and rerun on new matters. PATRON ships with built-in skills including a document reviewer, a devil's advocate mode, and a plain-language editor. Additional skills are available through the MateMatic Boutique catalogue.

Polish and EU Legal Source Connectors
Six open, MIT-licensed MCP connectors link PATRON to authoritative legal sources: SAOS (Polish court rulings including the Supreme Court, Constitutional Tribunal, and National Appeals Chamber), NSA (administrative courts), ISAP (statutes and official gazette), KRS (company register), EUR-Lex/CJEU (EU law and Court of Justice case law via live SPARQL), and an EU compliance connector covering GDPR, AI Act, DORA, NIS2, eIDAS 2.0, and CRA.