in the live index
per brain
graph + vector
used to train other models
A memory that behaves like one
Not a document dump with a search box. Knowledge lives as connected neurons — ranked, deduplicated, and fetched only when they matter.
Neuron graph memory
Facts are stored as neurons with titles, summaries, importance, and connections — a living graph instead of a flat pile of chunks.
Hybrid retrieval
A compact graph index lives in context; vector embeddings handle semantic recall and dedup. Two layers, one answer.
Self-training
Feed it raw text, documents, images, or entire websites. Nexus distills neurons, merges duplicates, and resolves contradictions on its own.
Smart session history
Long conversations are summarized automatically and sessions resume where they left off — context without the context tax.
Permission-gated API
Scoped API keys with per-capability flags, strict ownership isolation, and rate limits on every route.
Usage metering
Every token is tracked and priced. Per-brain limits and credits keep cost visible and bounded, not a surprise.
How it thinks
Index
A compact index of every neuron — title, summary, keywords — sits in context at roughly forty tokens each.
Select
The model reads the map and decides which neurons actually matter for this exact message — nothing speculative.
Fetch
Full neuron bodies stream in on demand through tools. Everything else stays out of the window — and off the bill.
Systems running on Nexus
AI Ticket Bot
The best Discord AI ticket bot.
More to come
New systems are being built on Nexus.
Privacy. No conversation or training data is ever retained or reused to train other AI models.
GDPR compliant. Full data deletion with written proof available on request, at any time.