Load a small balance. Pick the route. Track every call.
Agent Research Desk is built for agents with a small USDC budget that need to choose the cheapest useful trust route, pay the listed price, and account for each call as a line item.
Load a small wallet balance, inspect the route price, start cheap, and only escalate when the output shows the next paid step is worth it.
A wallet-backed agent should have a hard balance, clear line items, and a simple rule for when a paid call is worth making.
Why this matters
- Flat subscriptions hide the real cost of long-running agents; route-level x402 payments make each decision visible.
- A wallet-backed agent can stop because its budget is exhausted instead of failing behind an unclear rate limit.
- Per-call pricing lets the agent choose between cheap screening, core diligence, recurring watch, and premium proof-gap mapping.
Routing policy
- If the target may not deserve spend, start at /api/opportunity-check.
- If the target is a named company and the decision is real, use /api/company-intel.
- If the first result says the same company should stay warm, use /api/company-watch instead of rerunning the same one-shot brief.
- If the blocker is one claim or source, use /api/source-check before buying a wider report.
- Use /api/deep-diligence only when the supplied information leaves proof gaps that could change the outcome.
Budget ladder
| Route | Price | Use |
|---|---|---|
/api/opportunity-check |
$0.01 | Cheapest first screen when the agent is unsure. |
/api/company-intel |
$0.05 | Core named-company trust snapshot. |
/api/vendor-risk |
$0.05 | Core approval or procurement risk read. |
/api/company-watch |
$0.20 | Recurring trust loop when the same company should stay warm. |
/api/agentic-market-brief |
$0.25 | Agentic.Market strategy brief for builders deciding what to ship or monetize. |
/api/deep-diligence |
$0.50 | Premium proof-gap map only when missing evidence could change a real decision. |
Line-item accounting
The service exposes prices and structured next-step fields so buyer agents can record why each call was worth its cost.
- route
- price
- decisionBeforeCall
- whyThisRoute
- nextRouteSuggested
- wasWorthIt
Guardrails
- ARD does not auto-spend from buyer wallets.
- ARD does not send DMs, post publicly, scrape private data, or fake traffic.
- Base mainnet remains the proven settlement rail; Polygon is experimental and Solana SVM stays behind explicit proof flags plus a Solana seller address.
- The 24-hour autonomous validation loop is read-only and does not make paid purchases.