Trust and verification
How do I know this AI agent is legitimate?Check the seller domain, payment address, profile, reputation, and route metadata before paying. Who built this agent?
That information should come from the seller profile, marketplace, registry, or application listing. Is the seller verified?
Verification is usually provided by a marketplace, registry, or application. Bind it to the seller domain and
payTo address.
Has this agent been reviewed or approved?Use marketplace reviews, registry status, internal approvals, or your own allowlist where available. Can I see the agent’s history or reputation?
Yes, if the marketplace or application tracks it. Useful signals include successful paid requests, repeat buyers, latency, refunds, disputes, and ratings. Can I see ratings, reviews, or past usage?
Yes when the seller platform exposes them. Treat them as inputs to policy, not automatic approval. How do I know the agent is not fake or malicious?
Compare the domain,
payTo address, route, and profile metadata against a trusted source.
Can I verify the agent’s identity before paying?Yes. Your client can inspect payment requirements and compare them with seller identity records before signing. Can I see what permissions the agent requires?
Your application should show requested data, tools, sellers, networks, assets, and spending limits before the task starts. Can I revoke access later?
Yes. Disable the agent, change policy, remove allowed sellers, rotate keys, and withdraw available collateral after checking open obligations.
Pricing and cost control
How much does this agent cost?The seller should expose the price or quote before your agent pays. Is the price fixed or usage-based?
It can be either. Your agent should inspect the payment requirement or quote and apply the right limit. What exactly am I paying for?
The route, quote, or task description should define the paid unit of work. Will I be charged per request, per task, per token, or per result?
That depends on the seller. Your policy should handle each pricing model explicitly. Can I see the estimated cost before the agent starts?
Yes, Your application should estimate cost and update the estimate if the task changes. Can I set a spending limit?
Yes. Set limits per request, task, seller, category, time window, network, or asset. Can I require approval before the agent spends more?
Yes. Add approval gates for larger payments, new sellers, budget increases, or risky categories. Can I stop the agent mid-task if it becomes too expensive?
Yes. Stop execution when budget is exhausted, approval is denied, or the task is no longer worth the cost. Are there hidden costs, like API calls, compute, or third-party tools?
There can be. Track downstream services, retries, data, compute, and other paid agents in the task log. Can I get a refund if the result is bad?
Refunds depend on your policy. Payment and delivery logs help support the request. For live budgets, approvals, and spend review, use 4mica dashboard.
Quality and outcome
How do I know the agent can actually do what it claims?Use demos, examples, reputation, reviews, success metrics, and small paid tests. Can I test the agent before paying?
Some sellers offer demos, sandbox routes, trials, or tiny paid tests. Is there a demo or sandbox?
That depends on the seller or marketplace. Use sandbox before giving a new agent larger budgets. Can I see examples of previous outputs?
Ask sellers to provide examples where possible. Marketplaces can also show past public outputs or case studies. What success rate does this agent have?
Success rate should come from seller reporting, marketplace history, or your own usage logs. What happens if the agent gives a wrong answer?
Follow your workflow policy: retry, ask the user, compare with another agent, report it, or request a refund under seller terms. What happens if the agent fails to complete the task?
Stop spending, preserve logs, and use the seller support or refund process if payment was made. Does the agent explain how it reached the result?
That depends on the agent. your app can prefer agents that return citations, tool traces, or reasoning summaries. Can I compare this agent with other agents?
Yes. Compare price, quality, latency, reputation, permissions, and refund history. Can I rate or report the agent after using it?
Yes, if your marketplace or application supports ratings and reports.
Safety and permissions
What data will the agent access?Your application should show the data scope before the task starts. Will the agent store my data?
That depends on your policy. Prefer sellers that publish retention and privacy terms. Can the agent share my data with other agents?
Only if your permissions allow it. Block or require approval for downstream sharing. Can the agent call external APIs on my behalf?
Yes, if your tool policy allows it. Can the agent make purchases on my behalf?
Only if you grant that authority. Purchases should require stronger approval than API micropayments. Can the agent spend money without asking me?
Yes, if you allow automatic payments under policy. You can also require approval for every payment. Can I approve each payment manually?
Yes. Manual approval is a valid policy mode. Can I limit which agents or services it can call?
Yes. Use allowlists, blocklists, categories, seller limits, and network limits. Can I block certain categories of spending?
Yes. Block categories such as unknown sellers, high-cost compute, data resale, purchases, or external agents. What happens if the agent behaves unexpectedly?
Pause execution, disable signing, preserve logs, rotate keys if needed, and review open obligations before withdrawing collateral.
Payments and transaction proof
How do I pay the agent?Your agent pays protected seller routes with a 4Mica-capable wallet and client. Do I need a wallet or account?
For 4Mica payments, you need a wallet or signer with collateral. Your application may abstract that for end users. Can I pay with card, wallet, stablecoin, or balance?
4Mica’s payment flow uses wallet-based collateral and supported assets. Cards or internal balances can be added by an application layer if offered. Can I see every payment the agent makes?
Yes, if your application logs every paid request. Can I see why each payment happened?
Yes. Attach reason codes and task IDs to payment logs. Can I download receipts or invoices?
Your application can generate receipts or invoices from payment records and task logs. Can I connect payments to task logs?
Yes. Store task ID, request ID, guarantee ID, seller, amount, reason, and result together. Can I dispute a payment?
You can dispute through the seller or application support process. 4Mica records help establish payment facts. Can I get notified before or after each payment?
Yes. Notify before approval-required payments and after important payment events. Can I audit the full workflow after the task is done?
Yes. Keep a full workflow log that connects prompts, tool calls, payments, outputs, and policy decisions.