Summary: Tashi is worth cataloging not as just another DePIN, AI-agent network, or generic high-throughput consensus project, but as a coordination stack that explicitly splits local agreement, global infrastructure validation, and public-chain settlement into separate layers. The current whitepaper and docs make that decomposition unusually legible: Vertex provides leaderless DAG-based consensus inside application-specific meshnets, Lattice runs the global node market and reward-validation control plane through Orchestrators and Resource Nodes, and Arc bridges validated coordination outcomes to public chains only when payment, token issuance, or public finality is needed. That makes Tashi a useful comparison point for Streamr, Waku, replicated-service middleware, shared-sequencing systems, and crypto×AI coordination layers because the real control surfaces are not just latency claims. They are meshnet membership, Orchestrator power over validation and job routing, Reward Point issuance, staking-and-reputation policy, proprietary-versus-open consensus control, and when private coordination is translated into public-chain settlement.
What it does:
Provides application-specific peer-to-peer meshnets where robots, AI agents, games, or IoT systems reach fast local consensus through Vertex without publishing every event to a public chain
Uses a leaderless DAG plus virtual-voting design in Vertex to produce sub-100ms Byzantine-fault-tolerant ordering and a signed Proof of Coordination for each completed coordination session
Runs a global Lattice layer where Orchestrators handle discovery, routing, validation, reputation, failover injection, and reward co-signing while Resource Nodes provide proxying, relay capacity, and application-specific execution services
Treats Proofs of Coordination as the main portable output of the system: Orchestrators validate them for rewards and Arc can then bridge them to external chains for settlement, signaling, or token distribution
Pays for infrastructure through a USD- or USDC-denominated interface, issues Reward Points backed by Treasury reserves, and requires Resource Nodes to stake $TASHI for participation and slashing exposure
Positions most coordination as private and offchain by default, with Arc used only when outcomes need public-chain finality or tokenized settlement
Key claims:
The most reusable analytical split is Tashi’s three-layer architecture. Vertex handles local consensus, Lattice handles infrastructure coordination and economic policy, and Arc handles public-chain settlement. That is much more informative than filing it as a single AI coordination network.
Vertex’s docs stress leaderless DAG ordering, virtual voting, fairness, and Proofs of Coordination rather than ordinary block production. That makes Tashi closer to offchain replicated-service and coordination middleware than to a conventional smart-contract chain.
Lattice is where practical control concentrates. Orchestrators accept jobs, choose Resource Nodes, audit sessions, validate proofs, maintain reputation, inject failover, and co-sign reward proposals. Resource Nodes do the work, but Orchestrators own the higher-order routing and reward-validation layer.
The proof model is deliberately compact and session-scoped: supermajority signatures over agreed data hashes, optionally with more structure attached. That makes Proof of Coordination a reusable output object rather than a full public ledger.
The token and payment design is also revealing. Apps do not need TASHItousethenetwork;theycanpayviaaStripe−backedgatewayordirectlyinUSDC.OperatorsearnRewardPointsbackedat100pointsper1 and convert those to $TASHI. That pushes token demand toward operator extraction and staking rather than ordinary app usage.
Governance is materially centralized in the current materials. The Foundation controls protocol development and key parameters during the early phase, and Vertex is still proprietary though promised for later open-sourcing. That is an important maturity and governance caveat for any decentralized coordination reading.
Tashi clears the corpus bar because it makes a distinct control stack legible beneath the usual DePIN for AI/robots pitch: leaderless local consensus, global orchestration and reputation routing, proof-based reward validation, and optional public settlement only at the boundary.
Whitepaper: Tashi has a current official executive whitepaper and supporting docs rather than a separate PDF-heavy source packet. The main primary-source notes for this pass are in ../whitepapers/tashi-primary-sources-2026-05-15.md.