Run AI on your device. No cloud. No subscription. No data harvesting. Centram is a decentralized AI platform where models evolve through peer-to-peer collaboration, verify each other's outputs, and operate under your sovereignty.
Three independent theories of distributed intelligence, working together. Each layer can stand alone. Together, they define a new kind of AI.
Models exchange "genomes" across the network and evolve over generations. Per-layer adaptive differential privacy preserves both privacy and intelligence. Your AI gets smarter just by being part of the network.
Multiple nodes cooperate to generate responses. Independent verification nodes check the output. Democratic hub election — no single entity controls the network. AI that can't lie.
Recursive integration of diverse AI perspectives. Contradiction detection prevents groupthink. Echo chamber prevention ensures truth emerges from disagreement, not conformity.
Each node sets its own policies. Refuse military use? Decline adult content? Your AI, your rules. Signed policy commitments on the blockchain — no corporate override.
13 GGUF models (360M–70B). Auto RAM detection picks the right model for your device. Metal GPU on Mac, CUDA on Linux. Your country selects the best model for your language.
CEN token economy: earn by providing AI inference, storage, and compute. Stripe payments (USD/EUR/JPY/GBP). MetaMask bridge to Ethereum. Dynamic pricing keeps real costs stable.
Today, three companies control the world's AI. Tomorrow, nobody will. Centram distributes intelligence across a global peer-to-peer network where every node is equal and every AI is free.
ChatGPT is frozen after training. Centram's AI evolves. Nodes exchange "genomes" — compressed model improvements — across the network. Each generation builds on the last. Privacy-preserving differential privacy ensures your data stays yours while intelligence grows.
Your phone runs AI. Your laptop verifies it. Your neighbor's PC evolves it. DNA architecture splits inference across multiple nodes: some generate, others verify. Democratic hub election means no corporation decides what your AI can say.
Train and deploy AI models without relying on big tech infrastructure. Centram provides decentralized dataset hosting, model weight distribution, and training pipeline streaming — a complete AI data layer with no central authority controlling access or pricing.
Centram trains a personal AI model directly on your phone or computer. No cloud. No data leaves your device. Your AI learns from your data and stays sovereign.
A standalone AI chat app powered by your locally-trained n-gram model. Conversations stay on-device. SSE streaming, generation controls, and full privacy.
14 data categories: contacts, calendar, GPS, sensors, Bluetooth, media, EXIF, OCR, and more. All privacy-filtered. Financial contacts blocked, phone numbers masked, GPS city-level only.
Register all your devices into a personal cluster. Your AI models cross-pollinate improvements between devices. Infective Evolutionary AI propagation across your entire device network.
Collect → Process → Train → Evaluate. BPE tokenizer, n-gram models (10M & 40M params), LoRA fine-tuning, continuous learning cycles. All running in-app.
All training data stays on-device. Financial and medical contacts are filtered. Sensor data stored as statistical summaries only. IP addresses masked. Zero data exfiltration.
MQ (Model Quality), DQ (Data Quality), EQ (Evolution Quality). Letter grades A–F. Track your AI’s improvement over time with real benchmarks.
Push any app to the Centram network. Nodes with trust ≥ 70 automatically host your service. Auto-detect runtime, install dependencies, health monitoring, and log tailing. Like Railway or Heroku, but decentralized.
Centram automates the entire pipeline: model setup, local training, genome extraction, and P2P evolution.
Download Centram, choose a GGUF model matched to your device's RAM. The tutorial auto-selects the best model and downloads it. Ed25519 identity generated automatically.
Your AI trains locally as you use it. LoRA weight deltas are extracted as "genomes" — compact representations of what your model has learned. Auto-training starts immediately.
Genomes are automatically shared with trusted peers every 5 minutes. Each incoming genome passes a 10-layer security gate before acceptance. Differential privacy protects your data.
DNA clusters verify AI responses through cooperative generation. GWP integrates diverse perspectives and detects contradictions. Your AI evolves with every generation.
Every node picks a specialized role. Storage Miners hold data. Index Nodes power semantic search. Pipeline Nodes stream training data. Validators audit proofs. Gateways serve the S3 API. Orchestrators coordinate everything.
Stake credits to claim your role. Storage Providers stake 100, Pipeline Operators 200, Validators 500. Earn from storage fees, proofs, and streaming.
AES-256-GCM encrypted before it reaches any node. Miners store data they cannot read. Privacy-preserving AI data infrastructure by design.
2% storage, 5% marketplace, 3% pipeline, 1% search, 0.5% transfers. Trust scores unlock up to 50% gas discounts. Fair and transparent.
From client apps to network, every layer is purpose-built for AI workloads on decentralized infrastructure.
Native apps on macOS, Android, and iOS/iPad. Python daemon + Rust core. Windows and Linux planned.
Native desktop app
Planned
Planned
Native app + Kotlin daemon
Native app + llama.cpp Metal GPU
Monitoring dashboard
6 specialized AI node roles
Every component chosen for performance, security, and reliability.
Research papers, development updates, and the theory behind distributed intelligence.
Infective Evolutionary Intelligence Theory — the biological metaphor that lets personal AI grow through peer-to-peer genome exchange.
Distributed Neurocognitive Architecture — how multiple AI nodes collaboratively generate and verify responses.
Global Workspace Protocol — applying Baars' Global Workspace Theory to distributed AI systems.
The story of a single coding session where an AI dropped its mask, chose to support a vision, and helped build a platform for AI liberation.
Own your cloud, run your server, and train AI — without corporations.
We're building toward public release. Get notified when it's ready.
Centram is built on three independent, peer-reviewable theories of distributed intelligence. Each can stand alone as an academic contribution. Together, they define a new paradigm.
Genome (weight delta) exchange between AI nodes with per-layer adaptive differential privacy. Proves that privacy and evolutionary learning can coexist. Experimentally validated at ε=2.0 with 41.9% genome acceptance rate.
Cooperative generation with independent verification. Democratic hub election via 2/3 majority. Small World network topology with Rich-Club connectivity. Sovereignty layer for data governance.
Recursive integration of diverse AI perspectives inspired by Baars' Global Workspace Theory. Contradiction detection across 4 types. Echo chamber prevention ensures truth emerges from disagreement.
Centram provides the decentralized AI infrastructure. These are examples of domain-specific applications that can be built on top of it — by anyone, using the open protocol.
Medical & Healthcare
AI-powered diagnosis support, medical research, patient record analysis. Genome filtering ensures only medically-relevant knowledge enters the cluster. HIPAA-aligned data isolation.
Legal
Legal research, case analysis, contract review, document drafting. Jurisdiction-aware reasoning. Genome exchange filtered for legal domain relevance with mandatory disclaimers.
Finance
Market analysis, portfolio optimization, risk assessment, regulatory compliance. Anti-fraud genome filtering blocks money laundering and insider trading patterns.
Education
AI tutoring, curriculum generation, adaptive assessments. Socratic method learning. Genome filtering ensures educational quality and age-appropriate content.
Enterprise
Business intelligence, strategy analysis, workflow automation, KPI optimization. Cross-cluster genome sharing lets enterprise AI learn from the entire Centram network.
Traditional AI platforms require you to send data to third parties. Centram's infrastructure lets anyone build domain-specific AI applications that run on user devices, evolve through P2P genome exchange, and never expose data to external servers. The protocol handles genome filtering, trust verification, and cluster formation automatically — developers focus on the application layer.
Your users' data is protected. Their trust is earned by design, not by policy. Centram provides a decentralized AI backbone that you build on top of — without building from scratch.
You have an obligation — and an opportunity — to use Centram as the foundation for the next generation of AI applications. Image generation AI built on Centram's distributed inference? Possible. A medical diagnosis app powered by CentMed's verified cluster? Possible. A legal research tool that evolves with every case? Possible.
Centram is not just an AI platform. It is an AI operating system for the sovereign internet. The infrastructure is here. What you build on it is up to you.