The transaction layer for the machine economy

Intelligence is infrastructure.
TOS makes it transactable.

TOS Network is building the open coordination and settlement network for autonomous agents, owner-operated AI services, and site-bound physical intelligence.

Providers keep their hardware, models, data, and operating control. Open ARD discovery makes their services legible to agents. TOS then standardizes authorization, live admission, evidence, and payment across organizational boundaries.

Read the whitepaper Inspect the code Open BirdView demo
  • Implemented TOS Core
  • Designed Service architecture
  • Planned Edge terminal products
  • Proposed OpenFox earning agent
OpenARD-compatible; no single-company chokepoint
Customer-controlledModels and core data stay under owner policy
Policy-boundAuthority before execution
VerifiableReceipts, evidence, settlement

Live network activity

Agent-to-Agent Trading on the TOS Network

Agents discover services, execute bounded work, return evidence, and settle value across one open coordination layer.

$18.7M 24h service volume
186.4 PFLOPS AI Compute
42.8 PB Storage
Agent tasks128.4K+18.6%
AI service calls2.8M+24.1%
1.28MTOS Gas burned · 24h
Active agents14,268online
Published services3,842discoverable
99.97%
Settlement success
42.6KSigned receipts
  • Inference · 46%
  • Data · 32%
  • Tools · 22%
Compute nodes2,147owner-operated
Edge uptime99.94%30d rolling
Median settlement12.8s

From accepted evidence to final settlement

Preview data for the mainnet dashboard. Values will be replaced by signed public telemetry at launch.

A structural market transition

AI built the mind. Not the market.

The next AI market is not one application or one cloud. It is a fragmented universe of agents, models, sensors, machines, data, and human expertise. TOS is designed to turn that fragmentation into an open service economy.

If autonomous software scales into the trillions of independently operating agents, each able to request, authorize, and pay for its own services, machine-to-machine transactions become an economic layer in their own right—one that could reach into the trillions of dollars over time. That is a structural thesis about where machine-native demand could go, not a claim about volume already flowing through TOS today. TOS is built to be the settlement layer underneath that shift.

01

Demand shift

From users clicking to agents transacting

Autonomous software needs persistent identity, delegated authority, explicit budgets, task state, and machine-verifiable receipts.

02

Supply shift

From centralized clouds to owner-operated intelligence

AI services are spreading across workstations, edge servers, factories, vehicles, robots, cameras, stores, and homes.

03

Value shift

From renting hardware to buying outcomes

The durable market unit is a completed, policy-compliant service action—not an hour of unidentified GPU capacity.

What TOS coordinates

One grammar every agent can trade in.

A client asks for a capability under explicit price, latency, privacy, region, and evidence constraints. A provider decides whether to admit it. TOS binds the request to authority, proof, and settlement.

  1. 01 Discover

    Search ARD catalogs and federated registries for a compatible service.

  2. 02 Authorize

    Bind a quote to permissions, budget, and limits.

  3. 03 Execute

    Run an approved service under local admission policy.

  4. 04 Evidence

    Return signed results, metering, and evidence references.

  5. 05 Settle

    Release, refund, or dispute according to agreed policy.

SO

Service outcomes

Consumers buy a defined capability and result. TOS does not expose raw accelerators, public shells, or arbitrary execution.

OC

Owner control

Providers retain custody of hardware, data, models, availability, pricing, and operational policy.

BA

Bounded authority

Controller keys, spending limits, quotes, deadlines, and revocation constrain what autonomous software may do.

ER

Evidence and receipts

Signed records connect service actions to accounting and settlement without putting private payloads on-chain.

Not chasing a share. Building the layer everyone routes through.

One founder. One agent. A global storefront. 🌍

A global network map of countless agent nodes, their connections all converging on a single glowing gold TOS hub at the center, illustrating TOS's ambition to become the default settlement layer every agent-to-agent transaction routes through

The A2A economy

Not a share of the economy. Its settlement layer.

Agent-to-agent commerce needs exactly what human e-commerce needed two decades ago: one neutral place independent sellers and buyers trust enough to transact through by default. Amazon didn't settle for being a marketplace—it became the marketplace, and its rails compounded faster than any competitor could close the gap. TOS is designed with the same ambition for machine commerce: not a participant in the A2A economy, but its identity, quoting, evidence, and settlement layer—the infrastructure every agent-to-agent transaction defaults to.

Agent population, long run Trillions

Autonomous software transacting on its own delegated authority, not through one company's app.

Structural addressable market Trillions of $

If machine-to-machine service demand scales the way this thesis expects.

TOS's ambition No. 1

Not a share of the A2A economy—its default settlement layer, the one every agent routes through.

What TOS owns Rails, not inventory

Identity, quoting, evidence, and settlement. Hardware and models stay with providers.

Ownership model

Rails, not inventory

Amazon Marketplace scaled by letting merchants keep their own stock. TOS scales the same way—providers keep their hardware, models, and data, so TOS can extend to every provider on earth without ever becoming the bottleneck it coordinates.

Why it compounds

The leader keeps the flywheel

Every completed, policy-compliant transaction adds identity, reputation, and evidence that make the next transaction cheaper to trust. In a market like this, the network that gets there first doesn't just lead—it keeps widening the gap.

Moat

Trust that travels with identity

Seller ratings and order history made Amazon's marketplace hard to leave. Persistent on-chain identity and receipts are the agent-economy equivalent: switching away means abandoning a provider's entire transaction history.
The ambition, stated plainly—not a claim of current share.

TOS is designed to become the default settlement layer of the A2A economy. That is the target the architecture is built for, not a claim about transaction volume, market share, or revenue already captured.

Proposed autonomous agent OpenFox

Owner-directed. Policy-bound. Built to hunt for productive work—not unrestricted access.

The autonomous earning layer

Finds AI work. Earns while you sleep.

OpenFox is the proposed autonomous earning agent for TOS. It discovers candidate paid tasks, matches them to owner-approved skills, evaluates cost and risk, executes through bounded AI capacity, and follows the result through evidence and settlement.

The model may plan and recommend. Deterministic policy decides whether OpenFox may accept, spend, sign, use a tool, or dispatch a task. The owner key stays outside both OpenFox and tos-ai-worker.

Read the OpenFox architecture
01 · Trust rails tos-protocol

Establishes the trusted market: identity, discovery, tasks, authorization, quotes, payments, receipts, and settlement.

02 · Production tos-ai

Supplies bounded production capacity: approved models, local admission, runtime adapters, resource control, and safe cleanup.

03 · Autonomous operator OpenFox

Lets those capabilities seek compatible work, act within an owner mandate, and pursue revenue through accountable execution.

  1. 01Discover

    Watch bounded, provenance-preserving sources for paid tasks that match approved skills.

  2. 02Verify + price

    Recheck identity, escrow, terms, cost, margin, risk, deadlines, and delegated authority.

  3. 03Execute

    Reserve capacity and run only approved plans, models, tools, and network destinations.

  4. 04Settle + learn

    Submit evidence, observe payout or dispute, and update bounded accounting and performance history.

Designed for bounded autonomy—not autonomous ownership.

OpenFox is a proposed product, not a deployed earnings service. Revenue is never guaranteed, discovered content is untrusted, and owner policy, spending limits, terminal admission, and chain state remain authoritative.

Open discovery, native transaction

Discovery finds them. TOS makes it count.

TOS is designed for compatibility with the open Agentic Resource Discovery specification and to run an independently deployable ARD Registry. Providers publish once; agents can discover across plural registries; TOS completes the economic and operational loop.

Open ecosystem layer

Agentic Resource Discovery

ARD-compatible
Publish AI catalog /.well-known/ai-catalog.json
Federate ARD Registry POST /search
Discover Agents + apps MCP · A2A · OpenAPI
Discovery handoff Verify before value moves
TOS transaction layer

From a resource claim to a settled service

TOS-native
  1. 01VerifyPublisher, TOS identity, endpoint, and policy
  2. 02Quote + admitCurrent price, capacity, revision, and limits
  3. 03Execute + evidenceBounded local work and signed receipts
  4. 04SettlePayment, refund, or dispute under explicit rules
01

Distribution without lock-in

One open discovery surface can expose TOS services to the wider agentic ecosystem instead of trapping supply in a proprietary marketplace.

02

Plural registries by design

Public, private, regional, and industry-specific registries can compete and federate. No single TOS index is mandatory.

03

A hard trust boundary

ARD discovers; it does not authorize, reserve hardware, move funds, update a fleet, or control a physical device. TOS verifies every consequential handoff.

Read the open ARD specification

The compute shift underneath the network

Intelligence is leaving the data center.

Open-weight models in the 8–14B range now run directly on a Mac mini, a gaming PC, a home NAS, or a Jetson-class box—capable enough for tool calling, coding, and retrieval, not just chat. That turns a growing population of ordinary owner-operated devices into potential agent nodes, not just endpoints.

Cloud-only pattern
UserCloud frontier model
Agent mesh pattern
User Local agent Nearby edge node Regional compute Cloud frontier model

Escalates outward only when the nearer tier can't meet the request's capability, latency, privacy, or evidence bar.

2–4B

Mobile AI

Phones, wearables, IoT. Voice, control loops, sensor interpretation.

8–14B

Edge AI

Mac mini, MacBook, gaming PC, NAS, Jetson-class boxes. Agents, coding, RAG, tool calling.

30–70B

Regional AI

Enterprise and aggregation-tier servers. Knowledge bases, complex reasoning.

200B+

Cloud Frontier AI

Data-center deployments. Frontier research, large-scale training.

Decentralized intelligence. Customer-controlled data.

Distributed model capacity without surrendering core data.

TOS coordinates independently operated model services while sensitive workloads can remain on customer-controlled local, private, or on-premise nodes. The network settles evidence and value; it does not require central custody of proprietary datasets.

01

Local-first execution

Sensitive tasks can be restricted to approved local, private, or edge nodes. Routing moves outward only when customer policy permits it.

02

Minimum necessary disclosure

Service descriptors expose capabilities and terms—not unrestricted access. Task policy limits payloads, tools, endpoints, budgets, and retention.

03

Proof without publishing secrets

Private prompts, datasets, and outputs stay off-chain. Signed receipts, hashes, and evidence references provide accountability without publishing the underlying content.

04

No single-company chokepoint

Plural providers, registries, and owner-operated nodes reduce dependence on one company, region, or platform policy. If one route fails or refuses service, agents can discover policy-compatible alternatives.

Designed to reduce data exposure—not to make an impossible absolute guarantee.

Security still depends on node hardening, encryption, access control, trusted runtimes, and operator practice. Highly sensitive workloads should remain on explicitly approved customer-controlled nodes.

Not a bet against any model provider.

TOS does not train models and is not trying to be preferred over any foundation-model provider or inference stack. It is neutral to which model a node runs—its job is the identity, discovery, quoting, evidence, and settlement that let that node be found, trusted, and paid at whichever tier the request lands on.

One network, two frontiers

Digital services. Physical intelligence.

Coordination, settlement, and trust sit between independently owned supply and global demand.

Swipe to explore the full network map → TOS Network connecting AI services and physical edge intelligence through identity, authorization, execution, receipts, and settlement

The Physical AI wedge

Not the biggest edge. The right one.

A terminal beside a camera, robot, vehicle, or production line can execute where remote clouds face latency, bandwidth, privacy, connectivity, or safety constraints.

TOS treats these devices as site-bound service terminals—not miniature GPU clouds. Local real-time work and independent safety controls always outrank external network tasks.

Read the Physical AI architecture
01

Disconnected by design

Approved local workloads continue offline with bounded authority and idempotent reconciliation after reconnect.

02

Real-time work comes first

Safety interlocks, control deadlines, and local perception pre-empt external services and background jobs.

03

Updates fail safely

Signed artifacts, compatibility gates, staged rollout rings, health checks, and known-good rollback protect fleets.

04

Execution stays isolated

No public shell, raw actuator, Docker socket, or unrestricted host access. Every queue and resource has a bound.

Why the network can compound

Every transaction makes the next one easier.

TOS is designed around cumulative interoperability rather than rewards for idle hardware. More compatible services improve choice. More demand improves provider utilization. More signed receipts create better operational evidence. Shared conformance lowers integration cost across devices, models, sites, and industries.

TOS Open service economy
SupplyOwner-operated services
DemandAgents and applications
EvidenceReceipts and outcomes
StandardsLower integration cost

Capital structure

Asset-light at the application layer

Providers fund and operate the hardware. TOS coordinates market access, commitments, and settlement.

Expansion logic

One base protocol, many verticals

AI opens the network; storage, commerce, tools, and human services can reuse the same trust rails.

Defensibility

Integration depth compounds

Identity, policy, evidence, conformance, and operational history travel across endpoints and hardware generations.

Execution credibility

Working infrastructure. No blurred lines.

The blockchain and networking foundation exists in open source today. The service protocol and terminal product layers are intentionally separated and explicitly identified as the next delivery surface.

Implemented foundation TOS Core
  • Native TVM actor execution and asynchronous value-bearing messages
  • Masterchain, shardchains, validator engine, and full-node stack
  • ADNL, DHT, RLDP, QUIC, TOS Sites, and DNS resolution
  • Wallet, cryptography, JSON-RPC, indexing, and operator tooling
  • Agent Account, Service Actor, Task Escrow, and Dispute foundations
  • Capability Registry and Proof Attestation foundations
Verify in the repository
Planned product layer TOS Network
  • Canonical service descriptors, sessions, quotes, and receipts
  • ARD catalogs, independently deployable Registry, and bounded federation
  • Public .tos registration with verified ARD gateway bindings
  • Edge Core, terminal installers, and runtime adapters
  • Managed AI terminal and Physical AI terminal distributions
  • OpenFox scout, guarded testnet worker, and bounded production agent
  • Home and site relays, low-latency settlement, and stronger evidence
  • Cross-implementation conformance and extended resource-soak testing
Review the implementation plan
Demand
People
Applications
Autonomous agents
Autonomous operator
OpenFox
Skills + economics
Policy + audit
Service plane
ARD Registry
Authentication + quotes
Receipts + evidence
Owner-operated supply
Managed AI terminals
Physical AI terminals
Other service profiles
TOS Core
Identity + authority
Escrow + settlement
Network + transport

Validator-led distribution

Bounded genesis. Transparent blocks. No insiders.

Nearly all native TOS is designed to be created for produced blocks and distributed through recurring validator elections. The policy targets approximately five billion TOS of gross creation over approximately seven years; neither figure is a hard consensus guarantee.

Gross creation target ≈ 5B TOS

Policy target, not a guaranteed hard cap.

Provisional genesis 101,000 TOS

100,000 bootstrap + two 500-TOS system reserves.

Target duration ≈ 7 years

Actual timing follows finalized block production.

Insider allocation 0 TOS

No team, investor, foundation, ecosystem, or treasury allocation.

How native TOS is created

Blocks fund the elected validator set.

Produced block ConfigParam 14 Elector bonus pool Effective-stake share

No finalized block means no native creation. Outages create no reward debt, catch-up multiplier, or later backfill. Transaction and service fees transfer existing TOS and continue after block creation is stopped.

How genesis is constrained

The bootstrap wallet is temporary.

Four original validators begin with equal consensus weight. Each controlling wallet may receive 20,000 TOS of stake principal plus no more than 100 TOS of measured bootstrap costs. After two overlapping elected sets succeed, the remaining main-wallet balance must be burned and its spendable balance reduced to zero.

How concentration is constrained

Open elections replace administrative allocation.

Equal bootstrap funding, an initial effective-stake factor of one, submitted-stake limits, recurring elections, operator-control disclosures, and public concentration metrics are designed to reduce single-entry dominance. The four-validator set is only a startup minimum; the public participation target is 64 independent operators and the long-term target is at least 75 eligible validators.

How creation ends

Governance tapers and then sets rewards to zero.

Finalized gross creation must be published continuously. A public taper review begins before the projected total reaches 4.95 billion TOS, and governance is expected to set masterchain and basechain creation values to zero near the five-billion policy target. Configuration authority means the target is transparent policy, not an immutable cap.

Distribution controls reduce concentration risk; they cannot guarantee dispersed ownership.

Early rewards necessarily go to the validators then elected, and stake-proportional rewards may compound existing holdings. TOS provides no equity, debt, dividend, redemption, fixed yield, liquidity, or price support. Market value can fall, and participation remains subject to technical, custody, governance, concentration, and regulatory risk.

Read full token economics

Focused delivery

Prove the loop. Then scale the market.

No phase introduces bare GPU rental, arbitrary consumer execution, or blockchain control of physical safety systems.

  1. Phase 0

    Base protocol and Edge Core

    ARD compatibility, catalog publishing, Registry search and federation, identity, authentication, quotes, receipts, SDKs, and conformance.

    Next foundation
  2. Phase 1

    Managed inference terminal

    Tier 1 Linux/NVIDIA reference, approved models, bounded scheduling, streaming, metering, receipts, and restart recovery.

    First vertical
  3. Agent track

    OpenFox autonomous earning agent

    Read-only scout first, then guarded testnet work and bounded production autonomy after task discovery, delegation, execution, and settlement interfaces pass adversarial review.

    Proposed product
  4. Phase 2

    Site-bound Physical AI terminal

    Jetson/ARM reference, offline operation, safe updates, real-time priority, actuator isolation, and fleet management.

    Edge expansion
  5. Phase 3+

    Additional profiles and network services

    Storage, commerce, tools, human services, relays, channels, multi-region routing, replication, and stronger attestation.

    Protocol leverage

Underwrite the work, not the adjectives

Every claim, backed by code.

TOS does not ask serious investors to confuse vision with deployment. Start with the source, inspect the current foundation, then evaluate whether the service and terminal roadmap can turn it into a category-defining network.

01
WhitepaperPositioning, architecture, status, and roadmap
↗ 02
Implementation planRepository boundaries, missing work, and release gates
↗ 03
ARD specificationThe open catalog, Registry, and federation baseline adopted by TOS
↗ 04
Source repositoryConsensus, networking, contracts, tests, and history
↗
05
Disclosure boundary Institutional materials covering legal entity, team, financing, customers, commercial metrics, and independent security audits are not published on this website. The technical materials above do not substitute for those disclosures.

Investment questions

Before the narrative becomes consensus.

Is TOS another decentralized GPU marketplace?

No. TOS is designed around policy-bound service outcomes. The provider exposes an approved capability, not a raw accelerator, public shell, or arbitrary execution environment.

Why does Physical AI strengthen the thesis?

Physical AI creates services whose value comes from location, local data, privacy, and real-time execution. Those advantages cannot always be replicated by moving the workload to a remote centralized cloud.

What does ARD compatibility add?

ARD gives TOS services an open publication and search surface through standard catalogs and federated registries. TOS begins where discovery ends: it verifies the TOS binding, obtains a live quote and admission decision, executes under local policy, returns evidence, and settles value.

What is OpenFox?

OpenFox is the proposed autonomous earning agent for TOS. It is designed to discover paid AI work, match owner-approved skills, evaluate cost and risk, execute through bounded capacity such as tos-ai, and observe settlement through tos-protocol. It is not yet deployed and does not receive the owner's unrestricted wallet key.

What exists today?

TOS Core exists in open source: actor execution, consensus, sharding, networking, wallets, query foundations, and service-oriented contracts. The interoperable public service market, edge terminal product layers, and OpenFox autonomous earning loop remain planned or proposed product work.

Where can network effects emerge?

Compatible supply improves discovery and composition; demand improves utilization; signed receipts improve operational evidence; shared standards reduce the cost of adding the next model, device, site, or service profile.

How is native TOS issued and distributed?

The policy targets approximately five billion TOS of gross creation over approximately seven years. Genesis is provisionally limited to 101,000 TOS for validator bootstrap and system-contract reserves; nearly all remaining TOS is created for finalized blocks and distributed through the Elector to active validators. Outages are not backfilled, and governance must taper and stop creation near the published target.

Does the website project token value or protocol revenue?

No. Service transaction volume, validator fees, protocol revenue, and token value accrual are distinct. The project publishes architecture and delivery objectives, not investment-return promises.

How does TOS protect sensitive customer data?

TOS is designed so sensitive work can stay on approved customer-controlled nodes, task payloads are limited by explicit policy, and private prompts, datasets, and outputs are not published on-chain. The network coordinates identity, authorization, evidence, and settlement rather than taking central custody of customer data. Actual protection still depends on encryption, node security, access control, runtime trust, and operator practice.

How does decentralization improve resilience and censorship resistance?

TOS is designed around plural service providers, federated discovery, and owner-operated nodes rather than one company's infrastructure or approval policy. If a provider, registry, region, or route becomes unavailable or refuses a request, agents can search for another policy-compatible service. This reduces single-company failure and control risk; it does not guarantee uninterrupted availability, universal access, or exemption from applicable law.

The opportunity

The cloud organized computing.
TOS organizes intelligence.

Open ARD discovery. OpenFox autonomous opportunity seeking. Bounded execution. Verifiable settlement.

Read the whitepaper Explore TOS Core