Quick Start

Quick Start

Prerequisites

• Kubernetes cluster (KinD recommended)
• openverso-charts — Helm charts for Open5GS + UERANSIM
• Python 3.10+
• kubectl and helm configured against your cluster
• OpenRouter API key (or any OpenAI-compatible endpoint — also supports Chutes)

Install

git clone https://github.com/<you>/operaid.git
cd operaid
pip install -r requirements.txt

requirements.txt pins: openai>=1.12.0, requests, pyyaml, seaborn, matplotlib, pandas, numpy, scipy.

Configure

Set the openverso-charts path (used by nuke-deployment.sh / reset-deployment.sh):

export OPENVERSO_CHARTS_DIR=/path/to/openverso-charts

Set your API key based on provider:

export OPENROUTER_API_KEY="sk-..."     # for LLM_PROVIDER=openrouter (default)
# or
export CHUTES_API_TOKEN="..."          # for LLM_PROVIDER=chutes

Edit config.env to change defaults (model, max turns, timeouts, namespace, deployment profile). See Configuration Reference.

First Run

# Single experiment, S1 (NetworkPolicy fault)
./run_experiment.sh --api-key "$OPENROUTER_API_KEY" --model z-ai/glm-5 --scenario 1

# Same scenario but without tool access (ablation baseline)
./run_experiment.sh --api-key "$OPENROUTER_API_KEY" --model z-ai/glm-5 --scenario 1 --no-tools

# Full suite (YAML-driven matrix)
python3 run_suite.py --suite suites/tool_impact.yaml --api-key "$OPENROUTER_API_KEY"

Generate Paper Figures

python3 visualize_suite.py \
    --stats paper/figures/suite_statistics.json \
    --csv summary.csv \
    -o paper/figures

See Visualization for full options including pricing-aware cost analysis.