Reproduce

Run the Nora knowledge-agent harness checks from pinned public source. These development checks use synthetic documents and model doubles.

Prerequisites

To run the harness source, use Python 3.12, Node.js 22.22 or newer, and pnpm 12.8.1. The separate website build uses Node 24.

Run the source checks

Clone the public harness repository and check out the pinned commit, then run the commands below from the repository root:

git clone https://github.com/lm2000/nora-ai-knowledge-agent-harness-public.git
cd nora-ai-knowledge-agent-harness-public
git checkout 10bd370c711afabdfc7b23b516911c31ff58bcb4

From the repository root, create a Conda environment, install the locked Python dependencies, trust the mise configuration, install the frozen UI lockfile, and run validation:

conda create -n nora-ai python=3.12 pip
conda run -n nora-ai python -m pip install -c requirements.lock -e '.[server,gcs,dev]'
mise trust
pnpm --dir ui install --frozen-lockfile
make validate

These checks use synthetic documents and model doubles. They do not establish current production quality or live GCS cutover.

Expected results and limits

The expected result is that make validate completes without errors.

These development checks do not exercise real GCS bucket creation, credentials, or upload; local verification uses a FakeGCSClient double. Optional Docling conversion and additional client identities remain unverified. Follow the setup guide for the application and real retrieval smoke test.

For the dated evidence behind these checks, see Validation.

Reference

Run the source