Arkeion — the system of record
Optimises for auditability, versioning and query power. Time-travel, branches, SQL, full-text and vectors.
The database that fits in the case.
inro is an embedded key-value engine in pure Rust, built around one dominant criterion: the smallest possible consumption of disk, RAM and CPU.
From the Japanese 印籠 — the small, tightly fitted case that held a seal and its ink, carried on the belt, holding everything essential in the least possible space.
Footprint
Measured, not projected — phases 1 and 2 are closed. Each figure is fixed by a benchmark in the repository.
- 1,536 B
- a freshly created database — target was under 2 KB
- 34.14 B
- per 16-field log entry on disk — target was 40 or less
- 19.7 KB
- RAM per open database — target was under 32 KB
- 4 reads
- to open a database, the same at 30 KB and at 2 GB
Against Arkeion
inro — the local store
Optimises for footprint. Gives up SQL, time-travel and branching to get there. Thousands of instances per process, or one node with megabytes of RAM.
They coexist. Arkeion is where the record lives; inro is what the node writes to.
Where it runs
Edge and IoT
A few megabytes of RAM, slow CPUs, flash with limited write endurance, datasets under 100 MB, a single process.
Multi-tenant servers
Thousands of tiny databases open at once, one per tenant, where RAM per open database is the dominant cost.
Built for
quota's edge nodes.
Every quota node keeps its own row, one record per request with 16 fields, and a monthly aggregate — on a sliding retention window of two to fourteen days. Thousands of nodes, one database each: exactly the case where RAM per open database and the cost of pruning decide the bill.
Status
Name reserved. Phases 1 and 2 are closed and measured; the engine works and its numbers are in the repository. The crate is not published yet, and the next step is not another engine phase — it is running under quota's edge.
MIT or Apache-2.0, at your option.