Foundational AI · For the Law

Brelyne

A research-led developer of foundational AI models, purpose-trained for the legal domain — built to retrieve, reason over, and enrich legal text with the precision the work demands.

Legal retrieval · Document graphitization · Semantic infrastructure

Request Access See the Models
Scroll
The Thesis

General models
approximate the law.
Ours are trained on it.

BRELYNE is a research-focused developer of foundational artificial-intelligence models specialized for the legal domain. We do not bend general-purpose models toward legal work; we build models purpose-trained for legal text and the tasks that turn on every word — and we hold them to a single standard.

That standard is to make legal AI more effective, efficient, and scalable, and to set the benchmark for accuracy and efficiency in the field. Where others treat law as one more genre of text, we treat it as the discipline it is — citation, doctrine, structure, and consequence.

Everything we build begins with problems we have lived in legal practice and legal technology: from extracting a single date buried in a contract, to surfacing the one decisive citation hidden across thousands of cases.

The Models

Selected works

A small, deliberate family of models — each purpose-built for a stage of legal work: retrieval, enrichment, and the structure beneath both.

Retrieval Legal Embedding Model 01 / Vector search Retrieval Legal Reranking Model 02 / Relevance Enrichment Hierarchical Graphitization 03 / Knowledge graphs Infrastructure Semantic Chunking 04 / Open source
Model 01 · Retrieval

Legal embeddings

An embedding model trained to read law the way practitioners do. It turns a complex legal query into a dense vector that captures meaning, citation, and doctrine — so the right authority surfaces, not merely the right keywords. Where general-purpose embeddings flatten a statute and a sales contract into the same fuzzy notion of "document," ours hold the distinctions that decide a matter.

Paired with the reranker, it forms what we present as the most accurate legal retrieval pair available, ranking among the top performers on independent legal retrieval and embedding benchmarks. It is the difference between a search that returns plausibly related text and one that returns the case you needed.

Vector searchCitation-awareBenchmark-leading
Query → vector space → nearest authorities
Model 02 · Retrieval

Legal reranking

A reranking model that takes a field of candidate results and orders them by true legal relevance — weighing how directly a passage controls, distinguishes, or merely mentions the question at hand. Retrieval gets you close; reranking decides which of the close calls actually belongs at the top.

Together, embedding and reranking turn complex legal queries into highly relevant search results, supporting tasks such as searching large volumes of cases for a critical citation. They are designed as a system: recall you can trust, ordered by precision you can defend.

Relevance orderingCross-encoderPrecision@1
Candidates → reranked → controlling authority first
Model 03 · Enrichment

Graphitization

A hierarchical document-enrichment model that transforms unstructured documents of any length into structured knowledge graphs — laws, decisions, and contracts rendered as connected entities, clauses, parties, dates, and citations. It reads a document the way a careful associate would, then writes down the structure they would have held in their head.

It does this with sub-second latency, so enrichment is something you run across a corpus, not something you schedule overnight. The result is law you can traverse: query a clause, follow a citation, pull every party — as a graph rather than a wall of text.

Knowledge graphsAny lengthSub-second latency
Party Cite Date Court Clause Hold
Unstructured document → knowledge graph
Model 04 · Infrastructure

Semantic chunking

Beneath the models sits a semantic text-chunking algorithm we created and open-sourced. It splits documents where meaning actually breaks — at the seam between an argument and its exception, a clause and its proviso — rather than where a character count happens to fall. Good retrieval starts with good chunks; ours keep an idea intact.

It has become widely used infrastructure, adopted by several of the largest technology and software organizations, with more than two million monthly open-source downloads. It is the quiet layer that makes everything above it more accurate.

Open sourceSemantic boundaries2M+ downloads / mo
Split where meaning breaks, not where bytes do
By the numbers

What the work measures

2M+
Monthly open-source downloads
Semantic chunking algorithm
~26%
More accurate
vs. general-purpose equivalents
~30%
Faster inference
vs. the next-best alternative

Figures reported by BRELYNE. Our legal retrieval models rank among the top performers on independent legal retrieval and embedding benchmarks.

Deployment

Run it
on your terms.

The same models and APIs, reached three ways — hosted by us, procured through a major cloud marketplace, or running entirely inside your own walls. From fully managed to fully air-gapped, the model is identical; only the perimeter changes.

01

Platform

The BRELYNE platform. Hosted APIs you can call today, designed for production legal workloads.

  • Continuous availability
  • Minimal latency
  • Zero permanent data retention
02

Marketplace

A major cloud marketplace. Procure and deploy through the billing and security relationship you already trust.

  • Consolidated billing
  • Familiar procurement
  • Cloud-native scaling
03

Self-hosted

Private, air-gapped containers running in your own cloud tenancy or on your own hardware. Nothing leaves your environment.

  • Air-gapped containers
  • Your cloud or your hardware
  • Strict security & compliance
What the models do

Legal retrieval, embeddings, reranking, knowledge graphs, graphitization, semantic chunking, citation search, contract analysis, date extraction, case law, statutes, decisions, document enrichment, filings.

The work ranges from the deceptively simple to the genuinely hard. Pull every effective date from a stack of contracts. Find the one case that controls, somewhere in thousands. Map a filing into its parties, claims, and authorities — then ask it questions.

These are not hypotheticals. They are the tasks our team carried in legal practice and legal technology — the reason BRELYNE exists, and the measure we hold every model to. We build for the people who do this work: legal-technology companies and the law firms they serve.

On the horizon

A world of law,
structured.

A forthcoming offering: direct access to a proprietary repository of laws, decisions, contracts, and other legal documents from around the world — graphitized, retrievable, and ready for the models to reason over. The corpus and the models, built for each other.

Request early access
For developers

Build legal AI

Documentation, API credits, and the same models behind our platform — so you can build legal AI applications on retrieval, enrichment, and structure that were trained for the domain from the start. Call the embedding and reranking APIs, graphitize a document, or wire the chunker into your pipeline.

Reach the APIs however you deploy — hosted on the platform, through a cloud marketplace, or inside your own air-gapped containers. The interface is the same everywhere you run it.

Request API credits Request documentation access

Illustrative request & response. Citations shown are public landmark authorities.

Contact

Start here.

Tell us what you're building, or which matter you're trying to win. We'll route you to the right models, the right deployment, and the credits to begin.

We reply from contact@brelynetech.net. Your message opens in your mail client — nothing is stored here.

Message ready

Your email is composed and waiting in your mail client. Send it and we'll be in touch from contact@brelynetech.net.