The kernel binary of the Canonical Corpus Server. Source code is the product. The AI downloads it, understands it, builds what the use case demands.
Source code. Full exposure. An AI agent downloads the source, reads every line, understands the architecture, and builds exactly what the use case requires. No installer. No SaaS. No vendor lock-in. The source is the product.
The AI is the integrator. It reads the schemas, configures the corpora, stands up nodes, and deploys packet infrastructure tailored to the problem. The architecture is clean enough to understand completely — under 200KB of C, zero dependencies, every contract enforced at compile time.
A single C binary — the kernel. Packets arrive over the wire. The kernel writes them to memory-mapped ring buffers, manages their lifecycle across tiers, observes its own health using the same engine, and serves them back on demand.
Zero-copy writes into ring buffers. No serialization, no malloc. Frame arrives, memcpy, done.
Schema-driven corpora with automatic tiered lifecycle. Hot, warm, cold — configured once, managed forever.
The system writes its own telemetry using the same engine. Three timescales. The observer and the observed are one.
Search by corpus, device, time range. Cross-tier stitching. Hot-cache last record. The data is always there.
Every PPK node exposes three network surfaces.
Write. Data in, validated, stored, acknowledged.
Read. End-user data access. Search, query, retrieve.
Manage and adapt. Observation, control, configuration. Where the use case lives.
Same binary at every hop. Edges burn to gateways. Gateways burn to central. Hatch manages every connection. Users read from central.
No query parser. No garbage collector. No abstraction layers. The speed isn't an optimization. It's what happens when nothing is in the way.