Service outcomes
Consumers buy a defined capability and result. TOS does not expose raw accelerators, public shells, or arbitrary execution.
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.
-
01
Discover
Search ARD catalogs and federated registries for a compatible service.
-
02
Authorize
Bind a quote to permissions, budget, and limits.
-
03
Execute
Run an approved service under local admission policy.
-
04
Evidence
Return signed results, metering, and evidence references.
-
05
Settle
Release, refund, or dispute according to agreed policy.
Owner control
Providers retain custody of hardware, data, models, availability, pricing, and operational policy.
Bounded authority
Controller keys, spending limits, quotes, deadlines, and revocation constrain what autonomous software may do.
Evidence and receipts
Signed records connect service actions to accounting and settlement without putting private payloads on-chain.
Multiple payment rails
Commercial terms are quoted in supported stablecoins issued on TOS, with network fees paid in native TOS. A gateway may relay other request-time payment protocols, which is a transport choice, not a TOS Core consensus change.
One founder. One agent. A global storefront. 🌍
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.
Autonomous software transacting on its own delegated authority, not through one company's app.
If machine-to-machine service demand scales the way this thesis expects.
Not a share of the A2A economy, but its default settlement layer, the one every agent routes through.
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.
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
ARD-compatible
Agentic Resource Discovery
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
TOS-native
From a resource claim to a settled service
- 01VerifyPublisher, TOS identity, endpoint, and policy
- 02Quote + admitCurrent price, capacity, revision, and limits
- 03Execute + evidenceBounded local work and signed receipts
- 04SettlePayment, refund, or dispute under explicit rules
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.
Plural registries by design
Public, private, regional, and industry-specific registries can compete and federate. No single TOS index is mandatory.
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.
The next market underneath the network
AI has a wallet now. AI spends money.
That single fact opens a new track and a new revenue line at once. Agents don't wait for a human to check out. They pay per call, from their own funded Agent Account, inside limits their owner set in advance. Anything an owner can describe, price, and meter turns into something an AI buyer can find and pay for.
The new track
AI has a wallet. AI spends money.
Agent Wallets hold their own owner/controller keys, balance, and spending policy, so an agent completes a headless checkout, transacting on its own bounded authority instead of routing every payment back through a human.The new revenue line
Billed per call, not per license
Pay-per-call and metered pricing (quoted and settled in TOS or approved stablecoins, and compatible with live request-time protocols like x402) lets a dataset, model, or device earn from every agent request instead of sitting behind a one-time license or an ad-supported page built for humans.The new work
Owners package their own resources first
None of this happens automatically. Before a resource is agent-reachable, its owner has to describe it, claim it with a verifiable identity, price it, and let evidence build its track record, and TOS is designed to make each step verifiable on-chain, not just self-reported.Four steps turn a skill into a tradeable AI asset.
Turning a skill, dataset, model, or device into something an agent can find, trust, and pay for takes four concrete steps on TOS Network, not just switching on API access.
Describe it as a service
Turn the skill, dataset, model, or device into a machine-readable descriptor (what it does, under what terms, at what price) so agents can find it through ARD catalogs and registries.
Bind it to a verifiable identity
Attach the resource to a TOS account and capability grant, so authorship and ownership are provable on-chain, not just claimed in a listing an agent has no way to check.
Price it and open it to agents
Set a quote (per call or metered) and the access policy around it. That's the moment it stops being an informal skill and becomes something an agent can discover, authorize, and pay for.
Let evidence compound its value
Every completed call returns a signed receipt to the owner. That growing track record is what makes the same asset worth more the next time an agent calls it.
The asset that pays in the AI era isn't the raw content. It is the standardized access wrapped around it.
Controllable by owner policy, retrievable through an interface, callable by an agent, tied to a verifiable identity, and priced to settle automatically: that combination, not the underlying file or device alone, is what an AI economy can actually pay for.
Describing the target model, not a claim that TOS wallets are autonomously spending today.
Agent Account is implemented as a foundation in TOS Core and verifiable in the repository, but acceptance so far has run on throwaway localnets. A persistent public-testnet deployment has not happened yet. Autonomous spending specifically through OpenFox remains a proposed, undeployed capability, gated behind adversarial review of its task-discovery, delegation, execution, and settlement interfaces.