Skip to main content
NIOM learns about you over time. As you chat, it extracts key facts — your name, preferred languages, current projects — and stores them encrypted in ~/.niom/memory/brain/. These facts are injected into every conversation, so NIOM always has context about who you are.

​
How it works

You chat normally
  → Every ~6 messages, NIOM auto-extracts facts
    → Facts stored encrypted in brain/knowledge.enc
      → Next conversation loads brain context into system prompt
The extraction happens asynchronously after a thread is saved — it doesn’t slow down your conversation. NIOM uses the fast model (e.g., Groq Llama 3.3) to identify facts, keeping costs negligible.

​
What NIOM remembers

NIOM’s brain stores three types of information:
TypeExamplesHow it learns
Facts”User’s name is Arka”, “Works on the NIOM project”Auto-extracted from conversations
Preferenceslanguage → TypeScript, style → conciseAuto-extracted or manually set
PatternsRecurring workflows, common requestsAuto-detected over time

​
Managing your memory

​
Settings UI

Open Settings → Memory to:
  • View all stored facts and preferences
  • Add facts manually (e.g., “I prefer dark mode in all apps”)
  • Add preferences (key → value pairs)
  • Remove individual facts
  • Clear all brain memory

​
API

You can also manage memory programmatically: 
curl http://localhost:3001/memory/brain

​
Privacy

Brain data is encrypted at rest with AES-256-GCM, just like conversations and tasks. It never leaves your machine. Facts are only extracted from your local conversations — nothing is sent to any external service beyond your chosen LLM provider.

​
How it improves NIOM

With brain context, NIOM can:
  • Skip repetitive questions — it already knows your name, stack, and preferences
  • Tailor responses — if it knows you use TypeScript, it won’t suggest Python solutions
  • Build on past context — “continue working on the NIOM project” just works
  • Personalize tone — matching your preferred communication style over time