Isaacus

Embedding

Search legal text by meaning, not just words

Overview

Embedding

Embedding, or vectorization, is the process of converting content into sets of numbers that, when compared mathematically with each other, quantify how similar they are in meaning. Embeddings are used to power semantic search engines, text classification, and cluster analysis, as well as the retrieval component of retrieval-augmented generation (RAG) applications.

Use cases

  • Semantic search and document similarity

    Embeddings quantify semantic similarity — between queries and documents, or between documents themselves — so you can search cases, contracts, legislation, and other legal materials by meaning rather than exact keywords.

  • Retrieval for RAG

    Embeddings are the retrieval component of retrieval-augmented generation (RAG) applications: chunk your documents, embed the chunks and the query, retrieve the closest matches, and pass that context downstream. Use Kanon 2 Embedder as the retrieval layer.

  • Classification and clustering

    Use embeddings for text classification and cluster analysis — grouping similar clauses, matters, or documents when keyword overlap alone is not enough.

  • Benchmark-leading accuracy

    Ranks first on MLEB — the largest benchmark for legal text embedding models.

    Despite being many times smaller, Kanon 2 Embedder beats the performance of OpenAI’s Text Embedding 3 Large by 9% on MLEB, while simultaneously being 30% quicker than their fastest current-generation embedder, Text Embedding 3 Small.

    Thanks to its extreme parameter efficiency, Kanon 2 Embedder sets the new Pareto frontier in balancing inference time with legal information retrieval performance. In fact, it is the fastest of all commercial models currently on MLEB.

    The top 4 models on MLEB by NDCG@10 score

  • Privacy by default

    Your data is not used to train models by default.

    Kanon 2 Embedder is privacy and security friendly — none of your data is used to train our models by default — and can be self-hosted for enterprises with heightened security or reliability requirements.

    Kanon 2 Embedder is available for self-hosting via private, air-gapped containers on the AWS Marketplace.

  • Purpose-built over generic

    Pretrained on billions of legal tokens and finetuned on expert-reviewed query–document pairs.

    Improving on the original Kanon model’s 512-token context window, Kanon 2 Embedder supports up to 16,384 tokens. Like other modern text embedding models, it is Matryoshka-aware, supporting 1,024, 768, 512, and 256 dimensions, with a default of 1,792.

    The world’s most accurate legal embedding model is available through our embedding endpoint — with a quickstart guide to get you embedding your first query and document in minutes.

Ready to build with the world’s best legal embedding model?

Follow the quickstart guide or join the platform to get started.

Visit docsJoin platform