On November 12, 2026, Brave put an agentic payments proxy in front of the Brave Search API at search.agent.s.brave.app. Developers and software clients can buy web search, LLM Context, news, video, image, and three local-search endpoints at $0.005 per request ($5 per 1,000 queries) without creating an account or managing an API key. Unpaid requests return an HTTP 402 Payment Required challenge; the client signs a USDC transfer on Base (eip155:8453) with an x402 V2 client such as Stripe’s Purl CLI wallet, retries the call, and receives search results alongside a Base settlement receipt.

Removing credit cards and secret rotation speeds up command-line agent setup. Brave also faced an acute supply-side problem. One day earlier, on November 11, Brave published an account of its anti-scraping rollout explaining that Google’s google.com/goto link redirects pushed automated scrapers onto Brave’s independent index, triggering spikes of several thousand requests per second. Over the same period, Microsoft retired the Bing Search APIs, and Brave notes on its API portal that Google’s Custom Search JSON API shuts down on January 1, 2027. Charging $0.005 per call at an open 402 gate converts unmonetized crawler load into metered revenue while letting ephemeral agents skip signup friction.

Brave’s machine-readable proxy guide shows the real economic boundary between netted credit accounts and gross per-call payments in the three endpoints it excluded from x402.

In economist Perry Mehrling’s monetary framework (The Economics of Money and Banking, Lecture 2), money is a means of final settlement, whereas credit is a promise that delays settlement so parties can net obligations across time. A conventional API key backed by a credit card or prepaid balance works as an internal credit ledger. A developer funds an account once, runs ten thousand queries of varying sizes against the vendor’s database, and clears the net balance periodically through a card processor or monthly invoice. By design, search.agent.s.brave.app removes that netting layer. As Brave’s guide states, there is no session, no credit balance, and no top-up. Every request carries its own signed payment and settles individually through Brave’s facilitator.

Dropping the internal ledger creates two constraints that show up directly in Brave’s exclusion list.

First is a settlement price floor. Brave does not sell Autosuggest or Spellcheck over x402. Its guide gives the arithmetic plainly: those endpoints “cost less than the fee a facilitator takes for settling one payment, so they wait on settling many queries together as one payment.” Even on Base, verifying an EIP-3009 authorization and settling on-chain carries fixed per-transaction overhead. Inside a prepaid API ledger, metering a sub-cent autocomplete lookup adds near-zero billing cost. Under gross per-request settlement, any call priced below the facilitator’s settlement fee loses money if sold one at a time.

Second is upfront price certainty. Brave also keeps its Answers endpoint (/res/v1/chat/completions) off x402. On Brave’s standard API portal, Answers bills $4 per 1,000 requests plus $5 per million input and output tokens. Because token usage varies with prompt length and completion size, the final charge is unknown until generation finishes. An HTTP 402 challenge must quote an exact amount upfront in the Payment-Required header before work starts. Fixed-price, single-shot endpoints ($0.005 flat per search) fit that handshake cleanly; token-metered inference does not.

Even on the eight supported endpoints, replacing vendor credit with per-request crypto settlement redistributes unit costs, latency, and spend risk across the stack.

Brave currently covers Base gas fees “for a limited time” and operates its own settlement facilitator. That temporary gas subsidy keeps the buyer’s marginal cost at $0.005 today. Once gas coverage expires, or when third-party facilitators charge a spread on half-cent transfers, either the vendor absorbs those basis points out of its $5 CPM gross margin or the effective price per query climbs above standard prepaid rates. Standard API accounts also include $5 in free monthly credits and 50 queries per second on Search, terms absent from the stateless 402 proxy.

Stateless payments also change latency and spend governance. Each cold request takes two HTTP round-trips: an unpaid probe returning an empty-body 402 with a challenge valid for 300 seconds, followed by a signed retry. On a search product citing a 669-millisecond average latency in AIMultiple tests, uncached double round-trips add overhead to multi-step agent loops. Brave’s verify-before-settle sequence does protect callers from paying for broken infrastructure, since failed searches never settle and retrying an upstream 502 error costs nothing extra. What it cannot stop is a runaway agent loop. On a prepaid API account, teams rely on vendor rate limits, dashboard alerts, and hard monthly billing ceilings. With a funded Base wallet, a looping agent sending 500 valid HTTP 200 searches settles $2.50 on-chain with no card dispute or refund desk behind it, shifting spend control entirely onto client-side wallet funding and CLI caps such as purl --max-amount 5000.

Per-request HTTP 402 payments work well for zero-signup onboarding, sporadic CLI research, and cross-tool harnesses where provisioning an account costs more engineering time than a few hundred queries will ever cost in USDC. High-throughput production pipelines, sub-cent autocomplete lookups, and token-metered LLM endpoints still favor netted credit balances or batched payment channels that amortize settlement overhead across thousands of calls. Until x402 clients and facilitators support multi-query session netting, prepaid API keys stay the cheaper accounting layer at scale.