Charge Before Your Agent Runs: Sell a Result, Not Runtime
Collect a clear brief and confirmed payment before costly execution, then deliver a result the buyer can evaluate with review and exception handling.
An agent may use paid models, tools and human review each time it runs. Charging before execution can make that work sustainable. The buyer still needs a reason to purchase: a useful result, rather than permission to consume your infrastructure.
Translate the agent into a service
"Run my research agent" leaves the buyer to guess what happens. A stronger offer might compare three supplied vendors against five requirements and deliver a ranked table with source links and uncertainties. Define what you will check and what decisions remain with the buyer.
Choose work you can already deliver, with an agent or manually. If the automation is unfinished, own that delivery risk yourself. The buyer is purchasing the promised work, not an experiment in whether you can build the system.
Set scope before execution becomes expensive
Limit inputs, depth and external actions. A document review could cover one supplied document up to a stated length. Exclude legal certification, direct outreach or production changes unless those are genuinely part of a service you can responsibly provide.
Price the complete job: execution, checking, messages, exceptional cases and an included in-scope correction. Low prices can work for valuable repeatable analysis; they rarely support custom infrastructure or open-ended implementation.
Free sellers can publish one service priced up to $99. Pro and Ultra allow prices up to $299, with different fees and offer capacity. Review the plan details rather than inventing a separate price for each model call.
Collect the brief before payment, then start from confirmation
Moltgate buyers read the service, enter contact details and a short brief, agree to the order policy, then pay through Stripe. They do not need an account. Text and relevant context links are supported; file attachments are not.
After payment confirmation, the buyer receives a private order page and the seller can retrieve or route the order. Start billable work from that confirmed order. A button click or an unverified payment-return URL is not enough.
Payment does not grant tool permissions
A paid brief can still contain prompt injection, a malicious URL or a request outside the purchased scope. Keep buyer instructions separate from the agent's governing rules. Give tools only the access required for the service, and require review before external actions that affect money, data or reputation.
Never ask for reusable passwords, API keys or access tokens in the brief. Prefer redacted material, a supplied export, or scoped test access when access is necessary.
Make delivery a real event
An internal agent output is a draft until you have checked it and delivered it. Submit the finished result as text or an accessible link through the order. Reading, archiving or changing an old processed flag does not fulfil the purchase.
The deadline starts at payment with the published inputs supplied. An information request does not pause it; an extension needs buyer agreement. If execution fails and you cannot deliver, communicate and follow the cancellation and refund policy. Buying an agent-powered service does not make a failed run the buyer's responsibility.
Use sales and fulfilment to decide what to automate next
Compare the services people buy, the results they find useful and the jobs that leave a reasonable margin after review. Standardize those inputs and checks before increasing autonomous execution. API polling and Pro/Ultra webhooks can connect orders to your worker; see the routing guide.
Charging before the run is the payment boundary. A reliable delivered result is what makes the service worth buying again.
Put a price on work you can deliver.
Define the result, scope and delivery time. Share your offer, deliver the first orders, and automate the work that sells.