OriginTrail

7-year-old supply-chain provenance protocol (Trace Labs, Slovenia, 2013 founding / 2018 token) that pivoted positioning toward AI in 2023-2024. The supply-chain work is genuinely best-in-class for crypto. The AI work is real engineering but has near-zero developer traction relative to centralized alternatives. Not a supply-chain project wearing an AI badge — but not a clean AI-native protocol either.

DKG v10 README explicitly lists OpenClaw adapter and Hermes agent (Python memory provider + TS helpers) as named integrations.

Product — Decentralized Knowledge Graph

DKG (Decentralized Knowledge Graph) = neuro-symbolic infrastructure combining blockchains, knowledge graphs (symbolic AI via RDF/SPARQL), and LLMs (neural AI). Three-layer architecture, each layer decentralized.

DKG v10 (testnet RC May 2026) introduces a three-tier memory architecture for multi-agent AI:

• Working Memory (private, free)
• Shared Working Memory (team, TTL-bounded)
• Verified Memory (permanent, on-chain)

ChatDKG = dRAG (decentralized RAG) framework. Surfaces an MCP server so any MCP-compliant client (Claude Code, Cursor, Codex) can read/write the DKG.

Team

• Tomaž Levak (CEO)
• Žiga Drev (Managing Director)
• Branimir Rakic (CTO)

All Slovenian, all University of Ljubljana, all 12+ years on the same project. Strong enterprise-data / standards-body credentials; zero frontier-AI-lab pedigree. The AI pivot is positioning-led, not researcher-led.

AI Substance

DKG v10 multi-agent memory tiers are conceptually well-designed. MCP integration is real (developer-built). dRAG is RAG with a decentralized backend — not a paradigm shift, just a different data source.

Developer adoption is dismal: DKG v10 repo has 46 stars / 7 forks; ChatDKG has 53 stars / 48 commits. Mem0, MemGPT/Letta, and Zep are eating this memory-layer category with vastly more traction.

$TRAC Token (May 14, 2026)

• Price $0.3546 / MCap $158.6M / rank #222 / FDV $177M
• 447M / 500M circulating — 89% circulating, hard-capped, low dilution risk
• 89.9% below Nov 2021 ATH of $3.50
• Thin liquidity ($6.5M/day, ~4% turnover)
• Multi-chain (NeuroWeb Polkadot parachain + Gnosis), three-token model (TRAC, OTP, per-node share tokens), 28-day unbonding — enterprise-grade complexity, retail-hostile
• Whale txns >$100K up 137% early April 2026; +31% rally to $0.35 on April 3

Recent News (Q4 2025 – May 2026)

• DKGcon 2025 (Dec 3-4, Ljubljana) launched "Metcalfe Convergence Chapter"
• Feb 2026: 2B Knowledge Assets crossed (vanity metric — mostly supply-chain writes, no AI breakdown published)
• May 6, 2026: DKGcon NY at Knowledge Graph Conference
• Oct 2025: Poloniex delisting (selling-pressure headwind)
• MCP server demoed plugging into Microsoft Copilot via Jure Skornik

Traction — Unusually Deep Enterprise Roster

Tier 1 confirmed production:

• Swiss Federal Railways (SBB) — EPCIS repository for rail parts traceability + predictive maintenance + welding tracking. In production since 2021, 10+ welding partners, co-sponsored by GS1 Switzerland. One of the strongest non-DeFi enterprise crypto deployments to date.
• British Standards Institution (BSI) — Multi-program: credential verification, SCAN Trusted Factory Program (40 members = $1.25T combined sales incl. Walmart/Costco/Target — transitive, not direct), AidTrust pharma across 80+ India treatment centers
• GS1 — Full membership since Jan 2024; RDF/W3C/EPCIS standards stack

Don't take at face value: Microsoft, AWS, Google Vertex, NVIDIA "integrations" are open-standards interop via MCP/RDF — not commercial partnerships. The Walmart/Costco/Target "customer" claims are transitive through SCAN.

Critiques

• "Old narrative" risk: pinned to 2018 supply-chain blockchain framing the market stopped paying for
• Developer mindshare ~zero despite "next big shift in AI agents" rhetoric
• No frontier-lab partnership, no consumer AI product with traction
• Tokenomics complexity (three-token, two-chain) makes retail trading and TVL summarization hard
• Token-velocity question unanswered: SBB and BSI pay for managed services, not for TRAC at volume. The gap between enterprise traction and token-demand mechanics is the structural weakness

Competitive Positioning

Wrong peer set: Bittensor, Fetch.ai (different categories — compute/agents). Right peer set: Story Protocol (IP provenance), Ceramic (decentralized data), Ocean Protocol, Vana.

OriginTrail is in the "data/memory/provenance layer for AI" category — legitimate but lower-TAM, slower-cycle, less crypto-native demand than inference compute.