Repository navigation
Description
Summary
SNAPSHOT_ENABLED=true and SNAPSHOT_INTERVAL=3600 are correctly loaded via loadSnapshotConfig() and displayed in the boot log:
Git snapshots: ~/.agentmemory/snapshots (every 3600s)
However, no periodic timer ever calls mem::snapshot-create. The snapshot function is registered (registerSnapshotFunction), but unlike auto-forget, lesson-decay, insight-decay, consolidation, and recent-searches-sweep — which all have setInterval blocks in the startup section (~line 20545 of src-MOg2zKJs.mjs) — there is no corresponding setInterval for snapshot creation.
The result: snapshots are only created when manually triggered via the MCP memory_snapshot_create tool or POST /agentmemory/snapshot/create. If the daemon crashes between manual snapshots, all data since the last snapshot is lost.
Root cause (source investigation)
In src/index.ts (compiled: dist/src-MOg2zKJs.mjs), around line 20464:
const snapshotConfig = loadSnapshotConfig();
if (snapshotConfig.enabled) {
registerSnapshotFunction(sdk, kv, snapshotConfig.dir);
bootLog(`Git snapshots: ${snapshotConfig.dir} (every ${snapshotConfig.interval}s)`);
}The snapshotConfig.interval value is read and printed, but never passed to a setInterval. Compare with the other periodic tasks nearby (~line 20545+):
// auto-forget — HAS setInterval ✓
if (process.env.AUTO_FORGET_ENABLED !== "false") {
setInterval(async () => { await sdk.trigger({function_id: "mem::auto-forget", ...}); }, autoForgetIntervalMs).unref();
}
// lesson-decay — HAS setInterval ✓
if (process.env.LESSON_DECAY_ENABLED !== "false") {
setInterval(async () => { ... }, 864e5).unref();
}
// consolidation — HAS setInterval ✓
if (isConsolidationEnabled()) {
setInterval(async () => { ... }, consolidationIntervalMs).unref();
}
// snapshot — NO setInterval ✗
// (only registerSnapshotFunction + bootLog, no timer)Suggested fix
Add a setInterval for periodic snapshot creation, mirroring the pattern used by the other periodic tasks:
if (snapshotConfig.enabled) {
registerSnapshotFunction(sdk, kv, snapshotConfig.dir);
const snapshotIntervalMs = snapshotConfig.interval * 1000;
setInterval(async () => {
try {
await sdk.trigger({
function_id: "mem::snapshot-create",
payload: { message: "periodic snapshot" }
});
} catch {}
}, snapshotIntervalMs).unref();
bootLog(`Git snapshots: ${snapshotConfig.dir} (every ${snapshotConfig.interval}s)`);
}Environment
- agentmemory v0.9.27 (npm global)
- iii engine v0.11.2 (native)
- macOS (also affects all platforms — this is a code-level bug)
Context
This compounds with the data-loss-on-restart issues (#843, #913, #950). Even with the stop-order fix from #849 merged, a crash or kill -9 bypasses the shutdown handler entirely. Periodic snapshots are the safety net — but the net has a hole in it.
I have a local workaround (cron job calling POST /agentmemory/snapshot/create every 2h + auto-restore from latest snapshot in start.sh), which confirms the snapshot create/restore APIs work correctly. Happy to submit a PR for the setInterval fix if the maintainers agree this is the right approach.
Workaround
External cron job or start.sh script that periodically calls:
curl -X POST http://localhost:3111/agentmemory/snapshot/create \
-H "Content-Type: application/json" \
-d "{\"message\":\"periodic snapshot\"}"
Activity
merge: snapshot interval timer (rohitg00#1006, PR rohitg00#1039)
fix: env hydration, indexing, consolidation, connectors, hardening
fix: env hydration, indexing, consolidation lifecycle, connector acti…