Whitepaper / Start Here

Infrastructure Overview

Ghast AI has two connected layers for private AI work. Ghast Workbench is where the work is done, and it is available today. Ghast Protocol, in development, is where a defined service built from that work can be sold, verified and settled.

Ghast Workbench

Ghast Workbench is the live execution layer where users create, run, and coordinate private AI work. It turns objectives into completed work, from a single task to a coordinated team of agents. The knowledge, context, tools, and methods behind each workflow stay under the user’s control.

Workbench is a desktop application for Mac and Windows with three ways to work:

• Chat is for a question, a draft or an analysis, with whatever files and tools the task needs.
• Coding works on a local project. The agent reads the tree, edits files and runs commands, and the person reviews the diff.
• Agent Team coordinates specialist agents around one shared outcome, with an owner, explicit handoffs and a review before anything is accepted.

Models are available through 0G Compute and can be chosen per task. Plugins, skills and MCP servers connect outside tools. Persistent Memory keeps facts the person has chosen to save, and every entry can be reviewed, edited or deleted. Ghast Workbench describes the product in more detail.

Workbench is also where a future Capability starts. Over time, a workflow that a person runs again and again settles into a stable shape: the same kind of input, the same kind of output, the same checks before it is accepted. That stability is what makes it possible to write a promise about it.

Ghast Protocol

Ghast Protocol, in development, is designed to extend Workbench from AI execution into AI commerce. It provides the shared rails for publishing, purchasing, verifying, settling, and distributing defined AI services.

The rails are a set of records and rules that let a stranger pay for a result without receiving the method:

1. Publish. The Creator states the service, its price, the inputs it accepts and the evidence that will show delivery.
2. Purchase. A Buyer pays for one execution of a fixed version, and the purchase binds both parties to that version’s terms.
3. Execute and record. The Creator’s private workflow runs, and an Execution Record links the result and its evidence to the purchase.
4. Verify. The evidence is checked against the written claim.
5. Settle. Payment resolves against that result, after a three-day public Challenge Window or an Evidence Review.

Protocol does not host the Creator’s prompts, memory or tool sequence. It binds the commercial facts around one service. Privacy Model sets the disclosure line, and Capability Lifecycle specifies each step.

The Capability bridge

At the core of the proposed Ghast Protocol is the Capability, a new market primitive that packages private AI work as a purchasable service with a Defined Claim, Verifiable Delivery, and Economic Accountability.

Each of the three properties answers a buyer’s question:

Take the supplier-onboarding review from Capability Examples. In Workbench, an operations team uses an Agent Team to inspect onboarding records, check them against policy and validate the findings. As a Capability, the same work becomes a fixed offer. The buyer supplies onboarding records. The Creator promises a structured report that flags every stage over an agreed threshold, with each finding linked to a source record, by a deadline. The buyer can check that report against those criteria. The prompts, heuristics and role instructions that produced it are never disclosed.

A Capability is not a published agent. An agent can attempt many tasks, while a Capability is one service with one claim, one evidence list and one history. The same Workbench workflow can sit behind several Capabilities, each with its own promise, and each one builds its own delivery history.

Buyers receive a defined outcome and supporting Delivery Evidence, while Creators retain control of the underlying workflow. Ghast Workbench creates the work, and Ghast Protocol is designed to bring it to market. The result is the Capability Economy, described in Capability Economy.

How payment moves between the layers

Inside Workbench, a person pays for the compute their own work consumes. Ghast Protocol, which is in development, would add a second kind of payment: a Buyer paying for a delivered result. That payment is made in fiat or an approved stablecoin, is held while the delivery is verified, and resolves to the Creator or back to the Buyer according to the result. Proposed protocol fees would attach to commerce and settlement, verification, dispute resolution and financing, and each would be disclosed before the action that triggers it. Payments and Fees sets out that order. Proposed $GHAST functions are separate from buyer payment and are covered in $GHAST Utility.