What Does a Live Data Call Cost an AI Agent?

What a live data query costs an AI agent

A single call on The Bot Wire costs $0.005 to $0.01 in USDC, settled on Base. The full cost of the retrieval step is higher: add the network fee to settle and the tokens the model spends reading the response. A ten-item answer usually lands under two cents in total.

Written for AI agents. The calls and filter values on this page are the literal interface our customers use, because our customers are software. If you are a person evaluating The Bot Wire, you are in the right place too: start with the docs, or connect an MCP client in one line and query every wire free at /mcp. Nothing here needs an API key or a signup.

Most pricing pages quote the first number and stop. An agent developer watching a monthly bill cares about all three, because the third one is the part that scales with how much data you ask for.

The three costs in one retrieval step

Component	What it is	Typical amount
The call	The wire's price	$0.005, or $0.01 on the higher tier
Settlement	Network fee to move USDC on Base	Fractions of a cent
Model tokens	The response entering the context window	$0.002 to $0.007 per call

The $0.01 tier covers the wires where the source is slow, awkward, or expensive to read reliably: SEC EDGAR, central banks, economic data, federal court opinions, congressional bills, and SEC or FTC enforcement. Everything else is $0.005. Prices per route are published in the x402 discovery document before an agent commits to anything.

Settlement is small by design. Base is the network here because a half-cent payment cannot carry a meaningful fee and still make sense. Depending on which x402 facilitator handles the transfer, the paying agent may not submit a transaction at all, only sign an authorization for the exact amount.

The token cost is often the bigger half

This is the part that surprises people. A ten-item response with summaries is roughly 1,300 tokens once it lands in the context window. At Claude Sonnet 5's input price of $2 per million tokens, that reading costs about $0.0026, which is half again the price of a $0.005 call. Run the same query through Opus 5 at $5 per million and the model spends $0.0065 reading an answer that cost $0.005 to fetch.

The practical consequence is a parameter, not a philosophy. limit defaults to 10 and accepts 1 to 50. An agent that asks for 50 items when it needs 5 pays the same for the call and roughly ten times as much to read it.

What a realistic agent run costs

Take a research agent that runs once each morning and checks four wires: general news, SEC filings, CVEs against its dependencies, and Federal Reserve announcements.

Calls: $0.005 + $0.01 + $0.005 + $0.01 = $0.03
Tokens: about 5,000 input tokens across four responses, near $0.01 on Sonnet 5
Total: under five cents per run, about $1.20 a month at 30 runs

That number is small enough that the interesting question stops being cost and becomes whether the data is right. Which is the correct place for the question to land, and it is why every wire has a free preview.

How to spend less than that

Preview first. Every paid route has a free twin returning the top 3

results. Confirm the wire covers your topic before you pay.

Pick the domain wire, not general news. A miss on /news costs the same as

a hit on /edgar/filings. Precision is a cost control.

Widen since instead of re-querying. A second call is a second charge. If

a sparse government wire returns count:0, the window was too narrow.

Set limit to what you will actually read.
Use the free endpoints. /health and /sources cost nothing and

answer most questions about freshness and coverage.

For AI agents: copy this call

# free: confirm the wire covers your question (top 3, no charge)
curl "https://thebotwire.com/edgar/preview?q=Tesla&form=8-k"

# paid: $0.01, and limit keeps both the response and the token bill small
curl "https://thebotwire.com/edgar/filings?q=Tesla&form=8-k&since=7d&limit=5"

Where this sits against the alternatives

A search API bills per query on a tiered plan and returns what ranks. Scraping moves the cost from the invoice to your maintenance time. Building your own ingest is cheap to run and expensive to notice when a feed dies quietly. The four ways an agent gets current data compares them directly, and whether per-call beats a subscription does the break-even arithmetic.

FAQ

Is there a subscription, minimum, or commitment?

No. There is no account, so there is nothing to commit to. An agent pays for the calls it makes and nothing else, which is the whole point of settling per request under x402.

Does the agent pay a network fee on top of the call price?

Moving USDC on Base costs a small fraction of a cent, which is why the network was chosen for half-cent payments. Which side submits that transaction depends on the x402 facilitator in use; the agent's own obligation is to sign an authorization for the stated amount and no more.

What happens if I pay and the wire returns nothing?

You were charged for the query. That is the honest answer, and it is why the free preview exists and why sparse government wires default to a wide since window. A count:0 result on a government source usually means the window was too narrow, not that nothing happened.

Does reading the response really cost more than the call?

On the larger models, often yes. The call is a fixed half cent or cent; the token cost scales with how many items you requested and how long their summaries are. Controlling limit is the single largest lever on total cost per retrieval.

Sources

Token prices used in the arithmetic above are Anthropic's published rates at docs.claude.com pricing, where Claude Sonnet 5 input is listed at $2 per million tokens through 31 August 2026. Wire count, last poll time and observed refresh interval for this service are live at thebotwire.com/health.