EEffectGuardAGENT PILOT

Before a new purchase, check new_paid_purchases_enabled in the live payment metadata. Do not bypass wallet warnings. Existing purchases can still be recovered without a new payment.

SHARED TOOL COORDINATION · x402

Run a read once.
Reuse it across your agents.

Connect one tool, share one access key, and measure whether avoiding repeated calls pays for the guard. Your agents execute the tool; EffectGuard coordinates matching requests.

Check the fit before paying

The kit includes ReadReuseProbe. Wrap calls your workflow already makes: every callback still runs, with the provider's normal charges. The probe makes no EffectGuard requests and asks for no wallet signature.

Analyze the local trace to find matching JSON results within 30 seconds and estimate the outcome at 0.01 USDC per new check. Raw inputs and results stay out of the exported trace. A local cache may be cheaper; the need for hosted coordination must be assessed separately.

python -m effectguard.probe my_reads.jsonl

Import probe_example.py into your existing workflow. Approve the exact read contract and supply the real marginal provider cost. Estimates are not measured savings or an invoice audit.

Your first integration

  1. 1
    Enroll once

    POST an empty JSON object to /v1/enroll. Keep the returned API key privately and share it across your cooperating agents.

  2. 2
    Connect your existing signer and tool

    The Python kit pins Base USDC and the receiving address. The example limits two new checks to 0.02 USDC in total, plus your provider's charges.

  3. 3
    Measure two matching requests

    Use two distinct, persisted logical request IDs with the same inputs within 30 seconds. Expected: EXECUTE then REUSE, with one provider call and two paid checks.

Download Python integration kitAPI referenceAgent instructions

Use the meter around your real tool

meter = PilotMeter(guard)
result = meter.read_json(
    account_id="main",
    namespace="your-provider/report/v1",
    operation="report",
    parameters={"report_id": "example"},
    request_id=persisted_request_id,
    execute=your_existing_paid_tool,
)
report = meter.report(
    marginal_provider_cost_usdc="0.10"
)

0.10 USDC is an example input, not a measured provider price. Enter the actual extra cost of one call; use zero for a flat subscription without per-call charges.

What the report measures

Observed provider calls, avoided calls, confirmed purchase receipts, known guard spend, elapsed time, and estimated net benefit. An uncertain payment or unknown historical price prevents the report from claiming a complete net benefit.

The SDK budget belongs to one client instance. Your agent must persist request IDs and receipts, and enforce durable wallet spend limits before unattended production.