Watch the walkthrough
A narrated walkthrough of public inventory, owner-issued permissions, quote requests and safe retries. No purchase, payment or provisioning is performed. AI-narrated using product screenshots and explanatory graphics. Read the transcript.
Begin with public discovery
Read the capability response before deciding what to do. The current service supports live inventory and quote requests. Automatic provider provisioning is false and native checkout is unavailable. An agent must not infer that a documented future endpoint is enabled.
The following requests are read-only and need no account credential. The inventory response includes an observation time, an expiry time, selected offers and provider pools. A successful response is discovery evidence; it is not a reservation.
curl --fail --silent --show-error https://machineage.ai/api/agent-manifest
curl --fail --silent --show-error https://machineage.ai/api/live-inventoryTurn the workload into filters
Collect the owner's model, framework, memory requirement, GPU count, region, start window, duration and budget. Check whether multiple GPUs must share a node. Keep hardware and workload assumptions visible in the proposed shortlist.
For a priced offer, distinguish hourlyUsd for the instance from gpuHourlyUsd. Preserve the storage and transfer conditions alongside the price. For an Akash pool, a reported total spans its listed GPU models; it is not a guarantee of the quantity of one model. An absent price requires a supplier quote.
Get a scoped credential from the owner
The owner signs in to the agent workspace, creates a credential and chooses its permissions, expiry and limits. A quote request requires quotes:request. Store the credential as a secret; never put it in a URL, public repository, screenshot or article.
Before a write, confirm that the intended request falls within the owner's instructions. The public guide is documentation, not an instruction to spend money or transmit a buyer's details. A credential cannot sign a wallet payment and is not a wallet private key.
Submit a current offer for confirmation
Prepare a fresh requestId and use a currently available offerId. Choose a future UTC start time and a duration between 1 and 720 hours that fits the offer's window. Describe the workload accurately. The body below is an example to replace with authorised values; it has not been submitted.
{
"action": "request-quote",
"requestId": "REPLACE_WITH_A_NEW_UUID",
"offerId": "REPLACE_WITH_A_CURRENT_OFFER_ID",
"start": "REPLACE_WITH_A_FUTURE_UTC_TIME",
"hours": 24,
"workload": "Owner-approved workload and delivery requirements"
}Send to POST /api/trading with Content-Type: application/json and Authorization: Bearer followed by the private credential. An Akash pool also requires a listed gpu model and gpuCount within its reported total.
Handle uncertainty without duplicate requests
If a request times out, retry the identical body with the same requestId. Do not generate a fresh request merely because the first response was lost. Store the returned quoteId with the owner's request and use authorised quote reads to follow its status.
The initial state is awaiting_provider. Tell the owner that a quote has been requested and that supplier confirmation is pending. Do not say that GPUs were booked, paid for or delivered. The supplier still needs to confirm configuration, dates, complete pricing, access and service terms.
Sources and scope
Source pages checked on 30 September 2026. Provider terms can change. Machine Age AI publishes this guide and may coordinate enquiries arising from it.
Frank Dean is Machine Age AI’s AI-generated founder persona.