6-Layer Memory Architecture
LiveSession Logs
YYYY-MM-DD.md
On-demand load
Long-Term Brain
MEMORY.md
Loads at startup
Project Registry
projects.md
Loads at startup
Vector Memory
Semantic search
On-demand recall
Smart Loading
~4k token startup
80% savings
Neural Engine
INTELLIGENCE.md
Pattern analysis
Startup: L3 projects.md (1k) + L2 MEMORY.md (3k) = ~4k tokens · Session logs + Vector search = on-demand only · Neural engine = 24h cycle
Latest Intelligence
—Memory Token Budget
Daily Session Logs
Automatic daily archives of all agent interactions · memory/YYYY-MM-DD.md
No session entries yet.
Use agents in the dashboard — entries are automatically written here.
Write Manual Entry
MEMORY.md
Core intelligence file · Loaded at every agent startup
Append to Section
projects.md
Instant situational awareness · Loaded at startup (~1k tokens)
Search by Meaning
Retrieve relevant memories using natural language — not keywords
Enter a query to search semantic memory.
Memory grows with every agent interaction.
Store Memory Manually
Token-Efficient Startup
Load only what agents need. Reduce context from 20k+ → ~4k tokens.
Loaded at Startup
projects.md
Project registry with status and context
MEMORY.md
Strategic decisions, rules, core knowledge
Loaded On-Demand
Daily session logs
Only when historical context needed
Vector search results
Only when semantic recall needed
INTELLIGENCE.md summary
Injected as compressed briefing
Live Context Snapshot
Build Enriched Agent Prompt
INTELLIGENCE.md
Pattern analysis · Opportunity detection · Strategic insights · 24h cycle
Neural engine ready.
Click "Run Analysis Now" to generate intelligence from your memory base.
Auto-Curation Engine
Runs automatically every 3.5 days · Keeps MEMORY.md accurate and lean
Read Session Logs
Scans recent daily logs for high-value entries
Update MEMORY.md
Extracts decisions and curated insights
Update Projects
Refreshes project statuses and activity
Remove Outdated Info
Prunes stale data to keep context tight
Flush to Vector Store
Embeds important entries for semantic recall
Trigger Neural Run
Updates INTELLIGENCE.md with fresh analysis
Run Curation Now
Auto-runs every 3.5 days. Manual trigger available.
Write to Memory
Manually inject knowledge into any memory layer