Version 1.0 · September 2026. Ghast Workbench, the desktop workspace for private AI work, is available today. Ghast Protocol, which turns defined services built from that work into verifiable purchases, is in development. Start with The Problem, or install Workbench.
The most valuable AI workflows will never be public.
Ghast AI makes private expertise callable by agents. Ghast Workbench, the desktop workspace for Mac and Windows, is where that work is done today. Ghast Protocol, now in development, is the design for offering the result of that work as a service that a buyer can purchase and verify, without the method being handed over.
AI is moving from answers to work with real economic value. Yet the prompts, data, tools, and execution methods behind the best workflows often need to remain private. A security reviewer’s playbook, a trading desk’s routing logic and an operations team’s onboarding heuristics are worth something precisely because others do not have them. Publishing them destroys the value. Keeping them internal caps it.
The reason is structural. A market needs a unit small enough to price, standard enough to verify, and portable enough to trade. AI work has never had one. An agent is too coarse, a prompt is too fine, and a token counts model consumption rather than delivered outcome.
Ghast AI introduces the Capability: one defined AI service a buyer can purchase and verify, while the method behind it stays private. A Capability states what it will deliver, what evidence will show that it did, and what the Creator puts at stake if it does not.
The infrastructure in one view
Ghast AI is infrastructure for private AI work. Ghast Workbench is where teams build and run AI workflows, with the models, tools, and context under their own control. In the proposed design, Ghast Protocol would turn a proven workflow into a Capability.
The two layers run in sequence. In Workbench, a person uses Chat for a question or a draft, Coding for a local project, and Agent Team when several agents need to divide, hand off and review one outcome. Models are available through 0G Compute, and plugins, skills and Persistent Memory stay with the workspace. Once a workflow produces the same kind of result reliably, its owner could define it as a Capability. From then on, Protocol would handle the parts that a private workflow cannot provide by itself: a fixed promise, evidence, settlement and a public history.
Product status
Module
Status
Ghast Workbench
Live
Ghast Protocol
In development
Ghast Workbench is released for Mac and Windows. Everything in this whitepaper about purchasing, verifying, settling, challenging, bonding and financing a Capability describes Ghast Protocol, which is in development. The chapters that cover $GHAST describe proposed token functions. $GHAST has not been issued.
Ghast AI took first place at the 2026 0G APAC Hackathon, out of 293 submissions. Recognition and Traction records that result with its sources, and what it does and does not show.
How the whitepaper is organized
Start here.The Problem explains why internal use, full disclosure and opaque delivery each fail. Infrastructure Overview describes the two layers, and System Overview follows one piece of work from Workbench to a settled purchase.
Reference.Capability Examples works through three services, Token Allocation sets out how the proposed supply serves Capability Economy growth, and the Glossary holds the canonical definitions.
A Capability is one defined AI service, such as a security review of a pinned code revision, that a buyer could purchase and verify against a written claim while the Creator keeps the prompts, data and workflow private.
Can I use Ghast Protocol today?
Not yet. Ghast Protocol is in development. Ghast Workbench, the desktop workspace where Capabilities would be built, is available now for Mac and Windows.
Does this whitepaper offer a token?
No. Chapters that mention $GHAST describe proposed burn, lock, stake, and allocation functions. They are not an offer, and they do not state a price. $GHAST has not been issued, and no token function described here is live.
