tl;dr
We’re unveiling the world’s first legal AI API — the Isaacus API, a direct, programmatic interface to best-in-class legal AI capabilities.
The Isaacus API houses our flagship Kanon and Kanon Mini models, currently ranked as the most accurate legal AI classifiers of their size.
The API also boasts support for the Isaacus Query Language (IQL), the world’s first legal AI query language, that is, a query language designed specifically for analyzing legal documents with AI systems.
This is just the beginning — over the course of this year, we’ll be releasing universal extraction and embedding models, as well as a second generation of Kanon models.
A new way of doing legal AI
The Isaacus API is a new, first-of-its-kind programmatic interface for our legal AI models.
Right now, the API features a universal legal classification endpoint that plugs into our Kanon Universal Classifier and Kanon Universal Classifier Mini models.
Our universal classification models can take a legal document of any length alongside a statement about that document or a section of that document (for example, “This is a confidentiality clause”) and can classify whether or not the statement is supported by the document.
Unlike regular classification models, our universal classifiers don’t require any finetuning to be able to classify a document.
Using our universal classifiers, you can easily get started building your own legal AI apps and solutions for a range of problems, whether it be pulling out termination for convenience clauses from thousands of contracts like the Department of Government Efficiency (DOGE) recently had to do or trying to identify potentially unconscionable contractual clauses that might land you before the Australian Competition and Consumer Commission (ACCC) or whoever your local consumer rights regulator is.
In addition to the universal classification endpoint, our API also comes with support for the Isaacus Query Language (IQL), a first-of-its-kind legal AI query language that combines the elasticity of natural language with the precision of Boolean logic to help you extract the exact information you need from your legal documents.
Using IQL, you can invoke highly accurate, reusable query templates that abstract away the need for you to optimize your own idiosyncratic prompts for individual models. For example, to pull out termination for convenience clauses from a contract, all you need to do is run the query “{IS termination for convenience clause}”.
IQL also allows you to chain multiple statements together into more complex queries using Boolean and mathematical operators. For example, if one wanted to pull out potentially unreasonable contractual clauses by looking for confidentiality clauses that bind only a single side of a contract, they could use the query “{IS confidentiality clause} AND {IS unilateral clause}”.
What makes the Isaacus API so special
The Isaacus philosophy is simple: do one thing and do it exceptionally well, or don’t do anything at all.
Our API perfectly embodies that philosophy.
We’ve built our API with legal professionals in mind from day one, ensuring that our API is:
- Affordable: At a cost of only $1 per million tokens, you’d need to classify the US Constitution 160 times with our largest model to even spend a dollar, making our API especially well-suited for analyzing reams of legal data.
- Self-hostable: We’ve built secure, easily deployable containers that you can use to self-host our API on your own hardware, including inside air-gapped environments, and ensure your data never leaves your hands. Reach out to [email protected] to learn more.
- Performant: We take great pride in the fact that our legal AI models benchmark as the most accurate and efficient such models of their size, and we won’t give up that place easily.
- Batteries-included: We build our solutions with batteries included, whether that’s automatically sorting chunking out for you (using our immensely popular semchunk algorithm) or hand-optimizing query templates that take care of prompt engineering for you.
- Easy to use: We’ve invested heavily into ensuring our API is as intuitive, accessible and easy to use as possible, including by meticulously documenting its every nook and cranny.
Getting started in under 5 minutes
Getting started building your own legal AI apps and solutions with the Isaacus API takes less than 5 minutes.
First, create an Isaacus account, add a payment method, and generate an API key.
After that’s done, install our API client:
pip install isaacus
Next, use the code below to initialize your API client, download an example document to classify, and then print out any confidentiality clauses in the document. Make sure to swap in your actual API key.
from isaacus import Isaacus
# Initialize the API client.
client = Isaacus(api_key="PASTE_YOUR_API_KEY_HERE")
# Download an example document to classify (GitHub's terms of service).
tos = client.get(path="https://examples.isaacus.com/github-tos.md", cast_to=str)
# Classify the document using the Kanon Universal Classifier.
classification = client.classifications.universal.create(
model="kanon-universal-classifier",
query="This is a confidentiality clause.",
text=tos,
)
# Filter for clauses of the document that have a classification score over 0.5
# (a score above 0.5 indicates a positive classification).
clauses = filter(lambda clause: clause.score > 0.5, classification.chunks)
# Sort the clauses from highest to lowest classification score.
clauses = sorted(clauses, key=lambda clause: clause.score, reverse=True)
# Print the clauses.
for clause in clauses:
text = clause.text
score = clause.score
start, end = clause.start, clause.end
# Print the clause in the format:
# ---------- start char = {start} | end char = {end} | score = {score}% ----------
# {text}
print(
'-' * 10,
f'start char = {start:,} | end char = {end:,} | score = {score:.2%}',
'-' * 10,
'\n',
text,
end='\n\n',
)
And that’s it — you’ve just analyzed your first legal document with our API! We’ve only just scratched the surface of what’s possible, however. To learn more, check out our full quickstart guide.
What’s next?
This is just the beginning — over the course of this year, we’ll be expanding our offering to include:
- Universal extraction and embedding models and endpoints.
- Kanon 2 and Kanon 2 Ultra (🤫).
- Even more pre-optimized query templates.
- An API playground that makes it easy to interact with our models without the need for code.
You can stay updated on our progress by following us on LinkedIn. You can also reach out to us at [email protected] to answer any burning questions you might have or see how we can help you integrate Isaacus legal AI into your own workflows.
