Skip to main content
Velixar uses a 4-tier system to prioritize memories during recall. Higher-priority tiers are always returned first.

​
Tier overview

TierNamePurposeRetention
0PinnedCritical facts & preferencesPermanent until deleted
1SessionCurrent session contextExpires with session
2SemanticGeneral knowledge (default)Managed by relevance decay
3OrganizationShared across all users in a workspacePermanent
When you search memories, results from lower-numbered tiers are boosted: 
Final score = similarity_score × tier_boost
TierBoost
01.5×
11.2×
21.0×
31.1×
This means a tier-0 memory with 0.7 similarity outranks a tier-2 memory with 0.9 similarity.

​
Auto-tier routing

When using auto-extraction, Velixar automatically assigns tiers based on salience:
  • Salience ≥ 0.8 → Tier 0 (pinned) — high-confidence facts and preferences
  • Salience < 0.8 → Tier 2 (semantic) — general context
You can always override the tier manually via the API.

​
Examples

# Pin a critical preference (tier 0)
curl -X POST https://api.velixarai.com/v1/memory \
  -H "Authorization: Bearer vlx_your_key" \
  -H "Content-Type: application/json" \
  -d '{"content": "User is allergic to peanuts", "tier": 0, "type": "fact"}'

# Store session context (tier 1)
curl -X POST https://api.velixarai.com/v1/memory \
  -H "Authorization: Bearer vlx_your_key" \
  -H "Content-Type: application/json" \
  -d '{"content": "Currently debugging auth flow", "tier": 1, "type": "context"}'

# Org-wide knowledge (tier 3)
curl -X POST https://api.velixarai.com/v1/memory \
  -H "Authorization: Bearer vlx_your_key" \
  -H "Content-Type: application/json" \
  -d '{"content": "Company uses PostgreSQL for all services", "tier": 3, "type": "fact"}'