Current status
Ghast Workbench is available today. Protocol, Capability commerce, settlement and financing are proposed mechanisms, not services currently available to use.
In the current product, model requests and tools can send prompts, selected context or files to configured services. Data handling depends on the model, tools and sync settings you use. See Workbench’s current privacy boundaries.
What is a Capability?
One defined AI service a buyer can purchase and verify. It carries a Defined Claim, Verifiable Delivery, and Economic Accountability. The workflow and method behind it stay under the Creator's control.
Take a software security review. Sold as a Capability, it would promise a report on one commit within listed directories, with severity, location, evidence and a reproducible explanation for every issue. The buyer would check the report against that claim, and the review playbook would never leave the Creator. The glossary gives the short definition, and the Capability Definition chapter gives the full structure.
Origins and timeline
- : the original announcement below introduced Ghast’s proposed Capability Economy and the version 1.0 whitepaper.
- : the stable-client Team walkthrough recorded an Agent Team work item being assigned, reviewed, revised and accepted in the stable Workbench release, the kind of repeatable workflow a Capability is built from.
Capability Economy and the agentic economy
The agentic economy is a broader research discussion about agents representing people and businesses. Ghast’s proposal focuses on a defined AI service and the evidence used to evaluate its delivery. The two address different scopes.
| Question | Agentic economy research context | Ghast’s proposed Capability Economy |
|---|---|---|
| Main focus | Programmatic interactions, communication and market structure | A defined AI service, its promise and delivery record |
| Service checks | The research is broader than one service specification | Defined Claim and named delivery evidence |
| Resolution | Standards, market rules and governance are research questions | A three-day public challenge window and settlement against the evidence, in development |
Questions about the Capability Economy
Can I buy a Capability today?
Not yet. Ghast Protocol, which runs Capability purchases and settlement, is in development. Ghast Workbench, where the workflows behind Capabilities are built and run, is available today.
How does a buyer check delivery?
In the proposed design, the Capability names its success criteria and the evidence each order must produce before purchase. After delivery, that evidence would be checked against the Defined Claim, and anyone could challenge the result during a three-day public window. Read Evidence and Verification.
Is a Capability the same as an agent?
No. An agent can attempt many tasks. A Capability is one defined service with one claim, one price and one delivery history, and the same agent can sit behind several of them.
Can agents buy Capabilities, or only people?
Both, in the proposed design. An agent could compare Capabilities by their commitments, price, permissions and delivery history, invoke one within the scope and budget its owner approved, check the evidence and settle the order.
Does private mean all processing stays offline?
No. Model requests and tools in Workbench send selected context to the services you configure. The Workbench privacy explanation describes that boundary.
Sources and further reading
- Capability Economy glossary definition
- Capability Economy whitepaper chapter
- Defined Claims and Evidence and Verification
- agentic economy · research context and Ghast’s design question
- Microsoft Research: The Agentic Economy
Ghast AI is infrastructure for private AI work.
The work that never reaches a market
AI has stopped being about answers and started being about work with real economic value. But the workflows worth the most money are the ones nobody publishes. The prompts, the context, the tool chain, the execution logic: that is the entire advantage. Show it and it is gone.
So creators are left with three bad options. Keep the workflow internal and earn nothing outside the company. Open it up and watch it get copied. Or sell it as a black box and ask the buyer to take your word for it.
Creators need confidentiality. Buyers need verifiable delivery. Until now you could only have one.
Every market needs a unit
Before the shipping container, cargo moved by the crate, the barrel, the sack. Every load different, every port improvising. The container did not make ships any faster. It made shipping a market.
Every market runs on a unit like that: small enough to price, standard enough to verify, portable enough to trade. Equity has the share. Credit has the bond.
AI work has never had one. An agent is too coarse to price, because a single agent does a dozen different things and "it is good" is not a price. A prompt is too fine, because a prompt is an input rather than the work. A token measures how long the model thought and says nothing about what was delivered.
So the market never forms. The work has value. It has no form that can be sold.
The Capability
A Capability is one defined AI service a buyer can purchase and verify. One job, sold once.
It carries three things. A Defined Claim: the service promise, written and fixed before anyone pays. Verifiable Delivery: named evidence, measured against that promise after the work runs. Economic Accountability: the Creator locks $GHAST behind the Capability, and a confirmed breach slashes it.
What it does not carry is the method. The prompts, the data, the tools, and the execution strategy stay on the Creator's side of the boundary.
An execution desk will not open-source its strategy. It can still sell the execution: fill a defined order across approved venues, inside an agreed time window, under set risk limits. The buyer receives the fill, the price against benchmark, the fees, and the rule check. Nobody sees the strategy.
What a market does with a unit
Once a service is a unit, the things markets do become possible.
It can be priced. One claim, one price, one execution.
It can be trusted. A claim-met result is not final for three days. Anyone can challenge it by posting a bond. If a breach is confirmed, the buyer is refunded, the Creator's bond is slashed, and the challenger is rewarded. Payment, the Creator's capital, and the public record all resolve against the same claim.
It can travel. The same service can be reached through agents, applications, and APIs while staying one identity with one history.
It can compound. A Capability with a proven delivery record can raise growth capital against that record, repaid out of what it earns.
That is the Capability Economy: an open market where Capabilities are discovered, purchased, verified, and scaled, by humans and agents alike.
What exists today
Ghast Workbench is live. It is where individuals and teams build, run, and coordinate private AI workflows, with the models they choose, their own tools, memory that carries across tasks, and an Agent Team that can split one objective across specialized agents. It runs on macOS, Windows, and as a Chrome extension. The method never leaves the workspace.
Ghast AI took first place at the 2026 0G APAC Hackathon, out of 293 submissions from 1,145 builders.
Ghast Protocol is the commerce layer that will let Creators publish those workflows as Capabilities that people and agents can buy and invoke. It is in development.
Read the design
The Capability Economy whitepaper, version 1.0, covers the full design: claim structure, evidence and verification, the challenge and settlement path, the privacy boundary, $GHAST utility, and Capability financing.
Published .
About
Ghast AI is infrastructure for private AI work. Ghast Workbench is where the work gets done. Ghast Protocol turns it into services people can buy.