Inspector, Settings · Retention

Retention controls when, if ever, Neotoma drops or tiers data into cold storage. The default for a fresh install is keep everything, observations, source files, and timeline events are immutable by design and most operators rely on that for audit and provenance. The Retention panel exists for the cases where you do want to age data out.

inspector.neotoma.io/settings#retention

Inspector

Retention windows

Store	Window	After window	Size
Observations	keep all	-	84.2 MB
Source files	180 days	tier → cold	412.1 MB
Timeline events	keep all	-	12.6 MB
Conversation msgs	365 days	compact	31.8 MB
Sandbox writes	30 days	expire	0.4 MB

DB size

541 MB

hot + cold

If applied

−226 MB

next sweep

Last sweep

2d ago

manual run

Retention panel, per-store windows, cold-tier toggles, and projected size impact.

Stores

Observations, immutable row-per-write. Default is keep-all; any retention here is destructive and compounds, the reducer can no longer reconstruct prior states for affected entities.
Source files, the biggest store by bytes. Tier-to-cold-storage moves the bytes out of the SQLite blob columns to a configured cold path while preserving the metadata row, so observations still link to a known source even when the file is offline. Expire drops the bytes outright.
Timeline events, append-only event log. Default keep-all; expire is mostly used in CI / sandbox.
Conversation messages, chat turns. compact rewrites old turns to a summary row that preserves who-talked-to-whom-when without the full body.
Sandbox writes, only present on sandbox-environment instances. Defaults to a tight 30-day expiry so the public sandbox stays fresh.

Sweep semantics

Sweeps are opt-in, a nightly run can be enabled per store, but you can also kick a one-shot sweep from this panel and review the dry-run impact first.
tier → cold is reversible: the metadata row stays, the file path moves, and a re-ingest restores it.
expire and compact are not reversible, they drop bytes. Run them with backups in place.
Affected rows are recorded as a retention_sweep timeline event so the audit trail stays intact even after the source data is gone. See the Timeline for the running record.

Recommendations

Personal use, leave everything at keep-all; SQLite handles low-GB workloads comfortably.
Shared / server, tier source files older than 6-12 months to cold; keep observations and timeline events forever.
Sandbox / CI, expire everything aggressively; the dataset is synthetic.