Tiered Storage
advancedRetrieve from tiered memory (hot/warm/cold), view distribution, and monitor hit rates.
Tiered retrieve
POST
/v1/memory/tiered/retrieveRetrieve from the tiered store using text query, returning matched patterns with tier info.
querystringrequiredText query for tiered retrieval.
top_kintegerDefault: 5Number of results (1–20).
tierstring | nullFilter to specific tier: hot, warm, or cold. Returns from all tiers if omitted.
200Response
{
"results": [
{
"text": "Frequently accessed pattern",
"tier": "hot",
"confidence": 0.92,
"last_accessed": "2026-01-20T14:30:00Z"
},
{
"text": "Older pattern",
"tier": "warm",
"confidence": 0.78,
"last_accessed": "2026-01-10T09:00:00Z"
}
]
}Tier distribution
GET
/v1/memory/tiered/distributionGet current distribution of patterns across tiers.
200Response
{
"hot": 234,
"warm": 1089,
"cold": 3200,
"total": 4523
}Tiered stats
GET
/v1/memory/tiered/statsGet tiered memory statistics including hit rates per tier.
200Response
{
"hot_hit_rate": 0.89,
"warm_hit_rate": 0.45,
"cold_hit_rate": 0.12,
"promotions": 34,
"demotions": 12,
"total_accesses": 5670
}How tiers work
| Tier | Description | Access speed |
|---|---|---|
| Hot | Recently and frequently accessed patterns | Fastest |
| Warm | Moderately accessed patterns | Medium |
| Cold | Rarely accessed, archived patterns | Slowest |
Patterns are automatically promoted/demoted between tiers based on access frequency. The consolidation system also influences tier placement.
Next steps
- Memory Vectors — Low-level vector operations
- Performance — Cache and acceleration
- Consolidation — Sleep and merge operations