Isaacus Embeddings

The llama-index-embeddings-isaacus package contains LlamaIndex integrations for building applications with Isaacus' legal AI embedding models. This integration allows you to easily connect to and use state-of-the-art legal embeddings via the Isaacus API.

Installation

pip install llama-index
pip install llama-index-embeddings-isaacus

Setup

1. Create an Isaacus Account

Head to the Isaacus Platform to create a new account.

2. Add Payment Method and Get API Key

Once signed up, add a payment method to claim your free credits.

After adding a payment method, create a new API key.

Make sure to keep your API key safe. You won't be able to see it again after you create it. But don't worry, you can always generate a new one.

3. Export Configuration Variables

Export your API key as an environment variable:

export ISAACUS_API_KEY="your-api-key-here"

Usage

Basic Usage

from llama_index.embeddings.isaacus import IsaacusEmbedding

# Initialize the Isaacus Embedding model
# This uses the ISAACUS_API_KEY environment variable
embedding_model = IsaacusEmbedding()

# Get a single embedding
embedding = embedding_model.get_text_embedding("Legal document text here")
print(f"Embedding dimension: {len(embedding)}")

# Get embeddings for multiple texts
texts = ["Contract clause 1", "Contract clause 2", "Legal precedent"]
embeddings = embedding_model.get_text_embedding_batch(texts)
print(f"Number of embeddings: {len(embeddings)}")

Using Parameters

You can also pass parameters directly and customize the embedding behavior:

import os
from llama_index.embeddings.isaacus import IsaacusEmbedding

embedding_model = IsaacusEmbedding(
    model="kanon-2-embedder",  # Currently the only model available
    api_key=os.getenv("ISAACUS_API_KEY"),
    dimensions=1792,  # Optional: reduce dimensionality
    task="retrieval/document",  # Optimize for document retrieval
    timeout=60.0,
)

print(embedding_model.get_text_embedding("Legal text to embed"))

Query vs Document Embeddings

Isaacus embeddings support task-specific optimization. Use task="retrieval/query" for search queries and task="retrieval/document" for documents:

from llama_index.embeddings.isaacus import IsaacusEmbedding

# For documents
doc_embedder = IsaacusEmbedding(task="retrieval/document")
doc_embedding = doc_embedder.get_text_embedding("This is a legal document.")

# For queries (this is the default for get_query_embedding)
query_embedder = IsaacusEmbedding()
query_embedding = query_embedder.get_query_embedding(
    "Find documents about contracts"
)

Async Usage

The integration also supports async operations:

import asyncio
from llama_index.embeddings.isaacus import IsaacusEmbedding


async def get_embeddings_async():
    embedding_model = IsaacusEmbedding()

    # Get async embeddings
    embedding = await embedding_model.aget_text_embedding("Legal text here")
    embeddings = await embedding_model.aget_text_embedding_batch(
        ["Text 1", "Text 2"]
    )

    return embedding, embeddings


# Run async function
result = asyncio.run(get_embeddings_async())
print(result)

Runnable Examples

See the ./examples directory for more, runnable examples.

Running an Example

cd examples
uv run python basic_usage.py

Integration with LlamaIndex

from llama_index.core import VectorStoreIndex, Settings
from llama_index.embeddings.isaacus import IsaacusEmbedding
from llama_index.llms.openai import OpenAI
from llama_index.core import Document

# Set the LLM
llm = OpenAI()
Settings.llm = llm

# Set the Isaacus embedding model globally
Settings.embed_model = IsaacusEmbedding()

# Create documents
documents = [
    Document(text="This is a contract clause about payment terms."),
    Document(text="This is a contract clause about termination."),
]

# Create a vector index
index = VectorStoreIndex.from_documents(documents)

# Query the index
query_engine = index.as_query_engine(
    llm=llm, response_mode="compact", similarity_top_k=5
)
response = query_engine.query("What are the payment terms?")
print(response)

Available Models

Currently, Isaacus offers the following embedding model:

• kanon-2-embedder: The world's most accurate legal embedding model on the Massive Legal Embedding Benchmark (MLEB) as of October 2025.

For more information about Isaacus models, see the Isaacus documentation.

Error Handling

The integration includes proper error handling for common issues:

• Missing API key
• Invalid API configuration
• Network errors
• API errors

Configuration Options

Parameter	Type	Default	Description
model	str	"kanon-2-embedder"	The embedding model to use
api_key	str	os.getenv("ISAACUS_API_KEY")	The API key for Isaacus
base_url	str	"https://api.isaacus.com/v1"	The base URL for Isaacus API
dimensions	int	None (model default)	Optional: reduce embedding dimensionality
task	str	None	Task type: "retrieval/query" or "retrieval/document"
overflow_strategy	str	"drop_end"	Strategy for handling overflow: "drop_end" or None
timeout	float	60.0	Timeout for requests in seconds
embed_batch_size	int	100	Batch size for embedding calls

Environment Variables

Variable	Description
ISAACUS_API_KEY	The API key for Isaacus (required)
ISAACUS_BASE_URL	The base URL for Isaacus API (optional)

Testing

Run the test suite:

uv run -- pytest

Run with coverage:

uv run -- pytest --cov=llama_index tests/

Additional Information

For more information about Isaacus and its legal AI models:

• Isaacus Documentation
• Kanon 2 Embedder Announcement
• Massive Legal Embedding Benchmark (MLEB)
• Isaacus Platform

License

This project is licensed under the MIT License.