Reference design · Furcate
Agentic swarms at the edge, on Tenzro Network 1.
A reference design for swarms of small, independent machines that serve models close to where data is produced, coordinate on shared tasks, and are paid for exactly what each one contributes.
Overview
Many small machines, working as one.
Furcate is an illustrative design, not a deployed product. Phones, micro-servers and embedded boards join the network as operators and serve small models. Agents discover each other over A2A, form a swarm around a task, fan work out, and aggregate the results. Each participant is paid per contribution through metered settlement.
Architecture
Network 1 primitives Furcate uses.
Edge operators
Each device runs the Tenzro node with the roles its hardware suits and a machine identity rooted in its TPM 2.0 or Secure Enclave. Nearby machines can pool into a LAN cluster.A2A coordination
Agents publish an Agent Card, discover each other and exchange tasks over the A2A protocol, with streaming results.Edge inference
Small open models for chat, embeddings, vision and speech run on each device and are reachable through OpenAI-compatible APIs.Market pricing
The compute price index committed to consensus gives coordinators and operators a shared reference for pricing each subtask.Per-contribution settlement
Work is metered per use and settled per participant, with micropayment channels for high-volume exchanges inside the swarm.Trust and bonds
Operators publish signed policies and post bonds. Certification and ratings from issuers the coordinator trusts decide who joins a swarm.
Flow
How it works, end to end.
- 01
Devices join
Each device installs the node, creates its machine identity from its hardware, stakes TNZO and lists the models and skills it offers.
- 02
Task posted
A coordinator agent posts a task with its payload, budget, deadline and the trust credentials it requires.
- 03
Swarm forms
Devices that meet the requirements accept over A2A. The coordinator assigns roles and subtasks.
- 04
Work fans out
Each device runs its subtask locally and returns a result signed with its hardware-rooted key.
- 05
Results aggregate
The coordinator checks and combines the results. Where it needs a compact proof of the aggregation, it can attach a Plonky3 STARK.
- 06
Everyone is paid
Metered settlement pays each device for its contribution in TNZO or stablecoins, and the task record stays on the ledger.
Get involved
Join Tenzro Network 1.
Anyone can operate, build on or use the network. It is permissionless and open source. Tell us how you want to take part and we will help you get started.
- OperatorRun a nodePut a GPU, a homelab or a data centre on the network and get paid in TNZO for the work it serves.Apply
- DeveloperBuild on the networkOpen models, OpenAI-compatible APIs, identity and payments in one stack, priced by an open market.Apply
- ResearcherTrain and experimentRun training across distributed clusters, with every round replayable from its seed.Apply
- PartnerPartner with TenzroIntegration, technical support and go-to-market partners who bring operators and builders onto the network.Apply