We built the tool we couldn't find.
High-speed network infrastructure generates extraordinary amounts of data. The organisations that depend on it — broadcast engineers diagnosing a stream failure at 2am, carrier data teams trying to understand subscriber experience, quant developers reconstructing market microstructure — all face the same problem: the data exists on the wire, but by the time it's captured and stored, it isn't queryable.
Wirelake was built to close that gap. We started with the capture core — AF_XDP zero-copy ingestion at 100Gbps — and built outward from there, adding protocol-native decoders and an Apache Arrow/Parquet output layer designed from the ground up for the modern analytics stack. The result is a pipeline that takes raw network traffic and produces structured, Parquet-format data on local NVMe — ready for DuckDB, Spark, Grafana, or any tool your team already uses. No proprietary formats. No cloud dependency. No sampling.
How We Build
Our principles.
Write path independence
The local capture and write path never depends on network availability. Parquet is written to local NVMe first. Async replication and warm-tier offload are optional and never block the capture path.
No vendor lock-in
Wirelake output is standard Apache Parquet. You own your data, and you can query it with any tool that speaks Arrow — today and in ten years.
On-premises by design
For the industries Wirelake serves, data sovereignty is non-negotiable. Wirelake is deployed on your hardware, in your facility. This is a design requirement, not a configuration option.
Precision over approximation
Sampling is a compromise. Wirelake is built to capture every packet, decode every protocol, and timestamp every record with nanosecond precision. If the data is on the wire, it's in your lake.
Working on a hard network data problem?
We'd like to hear about it. We work directly with engineering and data teams to understand their environment before proposing anything.
Get in Touch →