Storage & Scale
Aionis Runtime uses SQLite as the local fact store. Substrate can be added as a durable evidence sidecar for backup, migration, preview context, audit, and stronger ordinary-memory recall.
Storage Layers
| Layer | Role |
|---|---|
| Runtime SQLite | Source of memory records, guide decisions, feedback, snapshots, and lifecycle state. |
| Substrate | Durable evidence mirror and recall sidecar for longer-lived inspection and migration. |
| Zvec | Optional ANN candidate index for larger recall workloads. |
| Agent host | Owns execution permissions, model calls, shell/browser access, and deployment boundary. |
Aionis keeps the governance boundary in the Runtime: recall sources can propose
candidates, but guide still decides use_now, inspect_before_use,
do_not_use, and rehydrate.
Deployment Shapes
| Shape | Use |
|---|---|
| Local Agent development | Start with npx aionis setup, run Runtime on loopback, connect SDK/MCP/plugin/AIFS. |
| Self-managed team host | Run Runtime behind your host boundary and keep tenant, scope, auth, and quotas explicit. |
| Durable local evidence | Add Substrate for backup, audit, migration, and ordinary-memory recall. |
| Larger candidate search | Add Zvec when memory volume makes ANN candidate generation useful. |
What To Measure
Large memory volumes should be measured for:
- candidate recall coverage,
- guide context size,
- embedding volume,
- Substrate mirror freshness,
- Zvec rebuild or incremental update latency,
- snapshot/report size,
- concurrent writes.
The recommended scaling path is:
Runtime SQLite -> Substrate sidecar -> optional Zvec candidates -> Runtime guideThat keeps Aionis’s core promise intact: stronger retrieval below, strict governance above, auditable memory influence throughout.
Last updated on