NOULI

Models / Text v1.0.0 · 31 October 2026

Nouli Clause

NDA evidence and entailment

Nouli Clause evaluates a fixed checklist of 17 NDA hypotheses. An Ettin-150M evidence retriever selects passages for an Ettin-1B decision encoder, which returns entailment, contradiction, or not-mentioned probabilities.

Comparison with Jev

Recorded comparison with Jev 1.14.0, dated 31 October 2026.

Nouli Clause: the published example, reference label, and recorded Jev comparison

Reference label: Supported. Nouli Clause selects Supported; Jev full-contract, calibrated selects Contradicted.

The excerpt is shown for the reader. Jev received the full NDA; Nouli used passages selected by its trained retriever.

Illustrative example selected after scoring; not a representative sample. Recorded evaluation · Methodology · Example details.

Model at a glance

Property Value
Parameters 1.178 billion
Input 512-token evidence windows; 1,280-token decision input
Tested runtime macOS ARM64 / CPU; Linux ARM64 / CUDA
License Apache-2.0; upstream notices apply

Quickstart with Nouli CLI

Requires Python 3.12, the HF CLI, and Nouli CLI 0.4.2. Hugging Face hosts the files; inference runs on your machine. The Python server loads the weights, and Nouli runs the supplied decision workflow. No hosted API key is required.

In the first terminal:

Terminal 1 · prepare and serve
hf download nouli-ai/nouli-clause --revision v1.0.0 --local-dir ./nouli-clause
cd nouli-clause
python3.12 -m venv .venv
source .venv/bin/activate
python -m pip install -r requirements.txt
python prepare.py --file examples/nda.txt
python serve.py --device cpu

Once the server reports ready, open a second terminal in nouli-clause:

Terminal 2 · run the workflow
nouli --config nouli/nouli.toml --store ./.nouli-store \
  workflow run nouli/example-workflow.json --json

The example uses CPU. On a configured CUDA host, start the server with --device cuda. Rerun prepare.py with your text or file before running another workflow.

Base models and training data

Training data: reuben256/contract-nli; 337 training contracts / 5,729 hypothesis decisions. The official source is Stanford ContractNLI, with CC-BY-4.0 attribution.

Trained on ContractNLI decision labels and human-annotated evidence spans. Source revisions and provenance.

Limitations

  • Supports the fixed 17-hypothesis NDA checklist, not arbitrary contract questions.
  • Retrieval errors can affect decisions. A retrieval miss does not establish that a statement is absent from the contract.

License and citation

Apache-2.0, with upstream notices and dataset attribution retained.

Nouli AI. Nouli Clause. Cite the published model version and commit.