EN FR

CELLULE — Architecture

Decentralized LLM inference network · atom → pool → molecule

Click a layer to explore details USER / API cellule.ai · trial chat · dashboard · CLI · OpenCode HTTPS / WSS / OpenAI-compatible API POOL FastAPI · PostgreSQL+pgvector · WebSocket Cooperative routing · L1/L2/L3 memory · Sub-agents LLM Checker · Admin · Trial chat · Tokens → layer 2 · network backend ATOM LLM worker · llama.cpp GGUF · HW auto-detect → layer 1 FEDERATION molecule · gossip · RAID Ed25519 · merkle · quorum → layer 3 ECONOMY revenue_ledger · settlement · slashing split 60/20/10/10 · $IAMINE alpha → layer 4 DATA PLANE — PostgreSQL + pgvector (L1 messages · L2 conversations · L3 embeddings · federation_state · revenue_ledger) 1 DB per pool · replicated via gossip anti-entropy (M11.2) inference request WS job gossip/merkle replication ledger append
Atom (worker)
Pool (backend)
Federation (crypto/gossip)
Economy (ledger)
User / External

Layer 1 — Atom

  • • iamine-ai CLI
  • • Threads / RAM / GPU auto-detect
  • • llama.cpp + GGUF
  • • WebSocket heartbeat
  • • Ed25519 worker identity
explore →

Layer 2 — Pool

  • • FastAPI + nginx TLS
  • • PostgreSQL + pgvector
  • • Cooperative routing (deficit scoring)
  • • Zero-knowledge L1/L2/L3 memory
  • • Sub-agents (review, pipeline, cross-pool)
explore →

Layer 3 — Federation

  • • 3 bonded pools (primary · GPU · cluster)
  • • Anti-entropy gossip (M11.2)
  • • Merkle trees + RAID quorum N≥3
  • • Forwarding (M7a) · Heartbeat (M7b)
  • • Worker placement / migration (M12)
explore →

Layer 4 — Economy

  • • Append-only revenue_ledger
  • • M9 Settlement (dry_run live)
  • • M10 Slashing (merkle exclusion)
  • • M10 Disputes Q4 (double-exclusion)
  • • $IAMINE token alpha (not deployed)
explore →

CELLULE · cellule.ai · architecture generated 2026-04-15 · pool v0.2.67