Overview
What is Waveguardclient?
Waveguardclient is a Python SDK for the WaveGuard API, a general-purpose anomaly detection service that uses GPU-accelerated wave physics instead of machine learning. It accepts any structured, numeric, or text data—server metrics, financial transactions, log files, sensor readings, time series—and returns anomaly scores, confidence levels, and explanations of which features triggered the alert. Designed for developers and AI agents who need zero-config anomaly detection without ML expertise.
How to use Waveguardclient?
Install via pip install WaveGuardClient, create a WaveGuard instance with your API key, then call scan() on your training and test data. For MCP integration with Claude Desktop, add a waveguard entry to your mcpServers config using uvx --from WaveGuardClient waveguard-mcp. All data types are auto-detected—no configuration needed.
Key features of Waveguardclient
- Stateless API: one call per detection, no training pipelines
- Works on JSON objects, arrays, text, and time series
- GPU-accelerated wave physics (no ML models, no drift)
- Returns anomaly scores, confidence, and per-feature explanations
- MCP server for Claude Desktop with three built-in tools
- Requires only 2+ training samples to define "normal"
Use cases of Waveguardclient
- DevOps server monitoring (CPU, memory, latency) to detect memory leaks or DDoS
- Financial fraud detection on transactions (amount, velocity, location)
- Security log analysis catching SQL injection, crypto miners, or privilege escalation
- IoT predictive maintenance from sensor readings (temperature, pressure, vibration)
- Time-series monitoring of metric windows for spikes or flatlines
FAQ from Waveguardclient
What are the dependencies to use Waveguardclient?
Only Python and the requests library. All physics computation runs server-side on GPU—you do not need a GPU locally.
How does Waveguardclient compare to traditional ML anomaly detectors?
It uses wave physics instead of machine learning, so there is no training step, no model drift, no hyperparameter tuning. Benchmarks show it achieves an average F1 of 0.76, competitive with scikit-learn’s IsolationForest (0.74), while requiring zero ML expertise.
Where does my data go when I send it to the API?
Data is encoded onto a lattice, processed on a GPU, and then torn down after each call. Nothing is stored on the server—the system is fully stateless.
What data types does the API support?
JSON objects, numeric arrays, text strings, and time-series windows are all auto-detected from the data shape. No manual type specification is required except for text (set encoder_type="text").
How do I authenticate with the API?
You provide an API key when creating the WaveGuard instance: WaveGuard(api_key="YOUR_KEY"). The key is passed in the request header.
