SpendWall
Custody-free · sits beside your payment loop

One wall for everything your agents spend.

Your agents pay OpenAI, Firecrawl, AWS, Stripe and x402 endpoints in the same afternoon. SpendWall puts all of it in one view, watches for the spike, and lets you freeze an agent in one click.

Start freeSee the union view

No card. The demo loads a real fleet with 30 days of spend.

Fleet spend · 30 days

$8,598

1 anomaly caught
$910$610$300$1.2k/day30 days agotoday+$1,720 · 2 daysrunaway scrape loop
  • LLM APIs
  • Data vendors
  • Cloud / metered
  • Payments
  • Crypto / x402

“An agent burned $4,000 over a weekend.”

The retry loop looked fine in every dashboard, because the money leaked across five vendors at once. No single bill looked wrong. The union did.

Five rails. One number.

Per-key caps and LLM gateways each see a slice. SpendWall is the only place that adds them up: model tokens, data pulls, metered cloud, card charges and x402 micropayments, side by side, per agent.

  • LLM APIs
  • Data vendors
  • Cloud / metered
  • Payments
  • Crypto / x402
Last 30 days, all railsNorthwind AI
$8,598one number
  • LLM APIs$3.5K
  • Data vendors$3K
  • Cloud / metered$550
  • Payments$1.2K
  • Crypto / x402$376

Three lines in. Zero latency added.

SpendWall enforces locally from a cached policy snapshot, so there is no authorize round-trip in your agent’s hot path. Freezing happens out of band.

  1. 1

    Wrap your fetch

    Swap fetch for sw.fetch. SpendWall classifies the rail, records the spend, and moves on.

  2. 2

    Enforce locally

    Policies evaluate against a snapshot the SDK already holds. No blocking call before a paid request, so no latency tax.

  3. 3

    Freeze out of band

    Revoke the agent from the dashboard or Slack. Its key stops working within seconds, without touching your payment path.

agent.ts
import { createSpendwall } from "@spendwall/sdk"const sw = createSpendwall({apiKey: process.env.SPENDWALL_KEY,agentId: "research-crawler",})// drop-in replacement for fetchconst res = await sw.fetch("https://api.openai.com/v1/chat/completions", opts)// spend shows up in the union view; a// freeze from the dashboard stops it here.

Watch, decide, stop.

Static anomaly rules you can reason about

Spend spikes, error bursts you're still paying for, request velocity, first-time endpoints, off-hours activity. Deterministic thresholds, not a black box.

Spend spikeError burstVelocityNew endpointOff-hours

Approve from Slack or your phone

Calls over a threshold wait for a yes. Approve or deny from wherever you are.

Policy as config

Caps, per-call limits, allowed rails and providers, per agent or per rail. Version it like code.

One-click freeze that actually stops spend

Freezing revokes the agent's key out of band. The SDK sees it on the next snapshot and blocks locally, so a rogue agent stops without you touching its payment path.

research-crawlerFrozen

SpendWall is not a budget cap.

A hard cap on one API key is free, and every gateway ships one. That was never the hard part. The hard part is seeing Stripe, Coinbase, your LLM bill and your metered cloud in the same picture, catching the pattern that no single bill shows, and stopping it before the weekend is over. That is the part SpendWall does.

Put a wall around your agents’ spend.

See two rails in one view this afternoon. Catch your first anomaly this week.

Start freeSee pricing