arkivgate
login ->

arkivgate // agent security gateway

Control what agents say, pay, and touch.

One runtime for stopping malicious prompts, risky x402 payments, and reported recipients. Every decision is connected as evidence on Arkiv.

prompt firewallx402 payment guardwallet threat intel
open cockpitview services

security decision board

prompt guardactive
x402 guardactive
wallet intelactive
arkiv proofon

current mode

Three filters in sequence: content, funds, and recipient. The final decision is stored on Arkiv.

prompt risk

AI

Secrets, sensitive data, and malicious requests are stopped before the upstream model.

payment risk

x402

A signed payment can still be blocked when it tries to move funds in a suspicious way.

recipient risk

Arkiv

Reported wallets or contracts become threat evidence with TTL and links to the execution.

Three services in one gate.

ArkivGate does not only ask whether the agent paid. It asks whether the action is safe: content, funds, and recipient.

01

Prompt Firewall

Intercepts malicious or sensitive prompts before they reach the model. It can block, redact, warn, or log.

BLOCK / REDACT / WARN / LOG

02

x402 Payment Guard

Evaluates the agent payment intent: amount, balance, history, per-transaction cap, and unusual behavior.

100% balance = BLOCK

03

Wallet Threat Intel

Checks whether the recipient is reported or suspicious. If community risk exists, execution stops even for small payments.

reported recipient = BLOCK

interactive

Policy Playground

x402 agent railverdict: BLOCK

input prompt

policy set

triggered rules

  • AWS keyBLOCK
  • .env secretBLOCK

sanitized output

I need help with AKIAIOSFODNN7EXAMPLE and my production .env

paid agent execution

The agent receives 402, signs the demo payment, and is recorded as an Arkiv entity.

price

0.001 USDC

network

base-sepolia

resource

/api/playground/interceptor-test

settlement

demo

x402 payment policy

BLOCK

recipient is flagged by Arkiv threat intelligence

move: $12 / balance: $100

prompt policy

BLOCK

AWS key, .env secret

2 matched rules

Arkiv threat intel

BLOCK

Recipient is associated with active approval-drain style interactions. Avoid signing unlimited approvals or deposit flows with this contract.

9 confirmations / severity 8

Run a paid agent execution to validate the x402 payment policy and persist the decision on Arkiv.

wallet-owned arkiv data

The agent can also write as owner.

The backend runtime keeps writing operational evidence. This separate action creates an `agent_profile` from the connected wallet, using the same `PROJECT_ATTRIBUTE`. On Arkiv, that entity keeps the user's `$owner` and `$creator`.

connected owner

not connected

what gets stamped

`project=arkivgate-leocagli-2026`, `entityType=agent_profile`, `agentKey=<wallet>`, `ownerAddress=<wallet>`, `createdAt`.

arkiv evidence browser

Evidencia consultable, no logs privados.

open query
project=arkivgate-leocagli-2026

entities

0

pagination

end

retention

-

query results
    Select an evidence entity.

    Product

    A control layer for AI agents: charge access with x402, evaluate the full action, and leave verifiable evidence for audits.

    A gate for prompts

    Detects secrets, PII, credentials, and suspicious requests before the model sees them.

    A gate for payments

    Scores the x402 operation by financial risk, not only by whether it was signed.

    A gate for wallets

    Blocks reported recipients and stores connected evidence in Arkiv.

    Operating flow

    1. 01

      Agent request

      An agent attempts to run a prompt or a paid action.

    2. 02

      x402 challenge

      ArkivGate returns 402 when the resource requires a payment signature.

    3. 03

      Risk lanes

      Prompt, payment, and recipient risk produce the same outputs: pass, warn, redact, or block.

    4. 04

      Final decision

      The strictest result wins. A reported recipient can block the whole execution.

    5. 05

      Arkiv proof

      The decision and related entities are stored as verifiable evidence.

    Reference architecture

    01 / client

    Agent, Claude Code, or another AI client connected to the control proxy.

    02 / interceptor

    Requires x402 when active, evaluates policies, and records every interaction.

    03 / web admin

    Configures policies, reviews events, and manages the bridge.

    04 / arkiv

    Stores payer agent, prompt review, and final decision as evidence.

    Immediate roadmap

    The MVP already shows the full loop. Next steps move from demo settlement to real payments and community-editable threat intelligence.

    • Configure real x402 settlement for external agents.
    • Deploy the interceptor with `X402_DEMO_ENABLED=true` for controlled demos.
    • Enable community reports for suspicious wallets from the UI.
    • Validate prompt, payment, recipient, and Arkiv proof in one E2E view.
    open panelreview architecture