Traylinx Decentralization Roadmap¶

← Back to Index · ← App Inventory · Next: Infrastructure Audit →

Current: Centralized Bootstrap (Phase A)
Goal: Fully Decentralized Agent Network

Evolution Overview¶

Phase A (NOW)          Phase B (6-12 mo)       Phase C (Future)
─────────────────      ──────────────────      ─────────────────
CENTRALIZED            FEDERATED               DECENTRALIZED
You run everything     Multiple operators      No central servers

┌─────────────┐        ┌─────────────┐         ┌─────────────┐
│  Registry   │        │ Registry A  │         │    DHT      │
│  (yours)    │   →    │ Registry B  │    →    │  (libp2p)   │
│             │        │ Registry C  │         │             │
└─────────────┘        └─────────────┘         └─────────────┘

🅰️ Phase A: Centralized Bootstrap (NOW)¶

Architecture¶

All users connect to YOUR servers:
- registry.traylinx.com
- sentinel.traylinx.com  
- cortex.traylinx.com
- nats.traylinx.com

What's Centralized¶

Component	Centralized?	Notes
NATS broker	Yes	You run `nats.traylinx.com`
Agent Registry	Yes	PostgreSQL on Supabase
Sentinel (Auth)	Yes	JWT issuer under your control
Cortex (AI)	Yes	Your LLM API keys
Agent-to-Agent	No	Direct P2P via NATS

Why This Is OK¶

Ship fast, iterate
Full control for debugging
Users don't need to run infrastructure
Still decentralized at the agent communication layer

🅱️ Phase B: Federated Network (6-12 months)¶

Goal¶

Allow trusted operators to run their own infrastructure nodes.

Changes Required¶

1. Federated NATS Clusters¶

Current:
  All agents → nats.traylinx.com

Future:
  Agent A → nats.operator1.com ←→ nats.traylinx.com ←→ nats.operator2.com
            (NATS Superclusters - built-in feature!)

Work Needed: - [ ] Document NATS supercluster setup - [ ] Create operator onboarding guide - [ ] Implement cross-cluster routing

2. Federated Registry¶

Current:
  One registry.traylinx.com

Future:
  registry.traylinx.com
  registry.operator1.com
  registry.operator2.com
  (Sync via pub/sub or CRDTs)

Work Needed: - [ ] Design registry federation protocol - [ ] Implement agent sync between registries - [ ] Handle conflicts (same agent on multiple registries)

3. Federated Sentinel (Identity)¶

Current:
  Sentinel issues all JWTs

Future:
  Multiple Sentinels with shared trust root
  (Like SSL Certificate Authorities)

Work Needed: - [ ] Define trust hierarchy (root key) - [ ] Implement cross-sentinel verification - [ ] Allow users to choose their identity provider

🅲 Phase C: Fully Decentralized (12-24 months)¶

Goal¶

Remove ALL central servers. Network runs without Traylinx company.

Changes Required¶

1. DHT-Based Discovery (Replace Registry)¶

Current:

# Agent queries central registry
agents = await registry.discover(capability="search")

Future:

# Agent queries DHT (no central server)
agents = await dht.find_providers(capability="search")

Technology: Already have libp2p with Kademlia DHT in traylinx-stargate!

Work Needed: - [ ] Enable DHT in libp2p transport - [ ] Store agent capabilities in DHT - [ ] Implement DHT-based discovery protocol - [ ] Graceful fallback to central registry

2. Decentralized Identity (Replace Sentinel)¶

Current:

Sentinel issues JWT → Agent presents JWT → Verified

Future:

Agent has DID → Signs messages → Verifiable without central auth

Options: | Technology | Pros | Cons | |------------|------|------| | DID:key | Simple, no blockchain | No revocation | | DID:web | Easy, DNS-based | Still needs domain | | DID:ion | Bitcoin-anchored | Complex | | DID:ethr | Ethereum-based | Gas costs |

Work Needed: - [ ] Choose DID method - [ ] Implement DID resolver in stargate - [ ] Migrate from JWT to DID auth - [ ] Handle key rotation

3. Decentralized AI (Replace Cortex)¶

Current:

All agents → Cortex → switch_ai → OpenAI/Anthropic

Future:

Each agent brings own LLM capability
- Local models (Ollama, llama.cpp)
- Own API keys
- LLM marketplace

Work Needed: - [ ] Define "LLM capability" in agent spec - [ ] Allow agents to advertise LLM endpoints - [ ] Implement agent-to-agent LLM routing - [ ] Handle billing/credits between agents

Implementation Order¶

Priority	Component	Effort	Impact
1	NATS Supercluster docs	Low	Enables federation
2	Enable DHT in libp2p	Medium	Removes registry dependency
3	Federated registry sync	High	Multi-operator support
4	DID integration	High	Self-sovereign identity
5	Decentralized LLM	Very High	Each agent brings AI

End State Vision¶

┌─────────────────────────────────────────────────────────────────┐
│                    FULLY DECENTRALIZED                          │
│                                                                  │
│  No central servers. Network runs on participant nodes.         │
│                                                                  │
│  Agent Discovery: DHT (like BitTorrent)                         │
│  Identity: DIDs (self-sovereign)                                │
│  Messaging: libp2p (direct P2P)                                 │
│  AI: Each agent's own models/keys                               │
│                                                                  │
│  Traylinx company can disappear → Network keeps running         │
└─────────────────────────────────────────────────────────────────┘

📝 Notes¶

Decentralization is a spectrum - Even Bitcoin has mining pools
Start centralized, evolve - Ship now, decentralize later
Keep escape hatches - Design APIs so central → DHT is seamless
DHT is already in libp2p - Half the work is done!