Tenzro
Reference design — Furcate

Agentic swarms at the edge, on Tenzro.

A reference design for decentralized physical AI at the edge — agentic swarms of distributed compute nodes coordinating tasks, sharing state, and settling on-chain through Tenzro Network's agent and settlement primitives.
Overview

What Furcate looks like, layered onto Tenzro Network.

Furcate is a hypothetical edge AI substrate where independent compute nodes — phones, micro-servers, embedded devices — host small models, expose inference, and coordinate via A2A. Swarms assemble around tasks, fan work out, aggregate results, and settle payment per contribution. Built on Tenzro's SwarmManager, AgentRuntime, A2A skill protocol, and micropayment channels.
Architecture

Primitives Furcate would use.

Swarm orchestration

SwarmManager creates swarms with persistent state. Broadcasts tasks. Tracks completion. Terminates atomically.

A2A coordination

Agents discover via Agent Card. Skills surface over JSON-RPC 2.0 with SSE streaming. 35 skills exposed at every node.

Edge inference

Small models (Qwen 3 0.6B, EmbeddingGemma 300M, Moonshine tiny) routed across edge providers via InferenceRouter.

Per-contribution settlement

Channels open per swarm participant. Ed25519-signed state updates. Aggregated and settled when swarm terminates.

Reputation routing

Provider reputation accumulates per-modality. InferenceRouter weighs reputation alongside price and latency.

AgentBond liability

Edge nodes post AgentBond. Counterparties file insurance claims for failed work. Bond slashed on misbehavior.

Pattern

From swarm formation to settled work.

Get started

Ship on the open network.