T
Open Ternary LLM
Watch a 10M-Parameter Brain Think
Tilelli (Tamazight for "freedom") is an open ternary language model running live inference on a single VPS with no GPU. Trained from absolute zero with {-1, 0, +1} weights. Every response is generated in real-time.
Protect
Truth
Respect
Privacy
Humility
10M
Parameters
Ternary
Weights
5
Pathways
0
GPU Required
This is research, not a product. You're seeing the raw output of an architecture experiment. The sidebar shows live router decisions and pathway activation as the model thinks.
Ready
Processed: 0
Tilelli Architecture
Live inference on CPU-only VPS
Model Specs
ArchitectureTilelli 5-Pathway
Parameters10M
Weight TypeTernary {-1, 0, +1}
Pre-trainingNone (from zero)
GPUNot required
LicenseApache-2.0
Router Activation (Live)
Local Conv
20%
State SSM
20%
Attention
20%
Wide Conv
20%
Dense FFN
20%
Last Response
Confidence--
Intent--
Mode--
Safety Laws
ProtectACTIVE
TruthACTIVE
RespectACTIVE
PrivacyACTIVE
HumilityACTIVE
Built from zero
tilelli.tech
tilelli.tech