FAQ

What is ProveXAgent?

ProveXAgent is a PulseChain-native layer on top of ProveX. ProveX is the settlement primitive (escrow + payment proof + release). ProveXAgent is the intelligence and permission layer: matching, tickets, recurrence, reputation, and templates. The product is not a chatbot. The product is a permissioned agent per wallet that can only plan and submit ProveX settlements inside a spend cap you set.

Deploy an agent once — it settles trustlessly forever via ProveX. Grok may explain the ticket. The agent never holds keys. ProveX moves the funds.

What is a permissioned agent per wallet?

When you register an agent (Atlas, a Recurring Strategy template, an A2A negotiator), you are not spawning a private copy of Grok. You are creating a named worker bound to your session — and, when the chain is live, to your wallet — with:

• A template (on-ramp, off-ramp, liquidity, A2A, …) and its tool allow-list.
• A max spend. The kernel refuses anything larger.
• A reputation floor (70). Counterparties below it are refused.
• Optional cadence (daily / weekly) so it can fire while you are offline.
• Stake in $PXA (vePXA when live) so misbehavior can be slashed.

That record is the product. Ten users share one Grok model. They do not share one agent. Each wallet’s agent has its own permissions, history, and stake.

How is this different from “an AI that trades for me”?

An AI that holds keys is a custodian. ProveXAgent forbids custody in the kernel: if a proof’s public signal says the agent held funds, verification fails closed. The seller signs createDeposit into ProveX Escrow. The buyer (or the agent posting on their behalf) submits fulfillIntent(paymentProof). The agent is a clerk with a badge, not a vault.

What is ProveX?

ProveX is a PulseChain P2P fiat-to-crypto settlement protocol. A seller locks crypto in escrow. A buyer pays fiat and produces a private proof of that payment. The contract releases. No centralized exchange sits in the middle. Core fee is 2%, 100% buy-and-burn of $PRVX. Open the live app at app.provex.com. Protocol site: provex.com.

Why does ProveX need agents?

ProveX is excellent at “did fiat move, then release.” It does not rank Zelle vs Wise offers, refuse a 42-score counterparty, run a weekly on-ramp, or compose multi-party proofs. Those jobs create volume ProveX would otherwise miss. Agents make ProveX the default rail for autonomous commerce instead of a manual tool.

Does ProveXAgent replace ProveX?

No. If the adapter cannot call the pinned Escrow and Orchestrator, nothing of value moves. We are not a fork, a wrapped ProveX, or a competing escrow.

Is ProveXAgent one Grok bot that every user shares?

No. There is one hosted Grok-4.5 model used as a narrator. Every user does not “log into the same bot.” They instruct their own agent record. Grok receives a short, stateless request: the template’s system prompt plus this instruction. It returns 2–4 sentences. It cannot open escrow, sign, or see other users’ chats.

Does each user spawn their own Grok?

No, and they should not. A private Grok per wallet would be slow, expensive, and a worse security model (persistent model memory is where keys and payment details leak). Isolation belongs in permissions, stake, and the kernel — not in a dedicated LLM process.

How is Grok connected to a user?

It isn’t bound as an account. The Console calls a server function with the prompt. If an xAI key is present, Grok narrates. If not, the local plan text is used and the ticket still renders. Chat in this demo lives in your browser (localStorage), not in Grok.

How does this scale?

Scale is request volume, not bot count. Ten thousand instructs are ten thousand short HTTP completions against one model, plus local kernel checks. Recurring agents do not keep Grok online; a scheduler fires the same instruct path later. The live bottleneck will be PulseChain RPC and ProveX escrow, not Grok.

What if Grok is down or rate-limited?

Settlement continues on the local plan. Narration is optional. Funds never wait on a language model.

What happens when I type in the Console?

• The kernel sanitizes the prompt (no scripts, length cap 500 characters).
• A deterministic planner picks a template (on-ramp, off-ramp, deploy, stake…).
• It selects an active agent that matches that template, or the first active agent.
• Gate: pause, dust ($10), max size, asset allow-list, agent max-spend, rate limit, reputation.
• A pre-trade ticket shows ProveX 2%, Agent 0.5/0.5, you-receive, seller-locks, 2.4% $PRVX pressure.
• The offer book scores inventory. Rows below floor 70 are refused.
• You confirm. The demo steps a 12-step rail. Live, that becomes on-chain calls.

What are the on-chain calls when this is live?

Seller: Escrow createDeposit at 0x0B42…f11a. Buyer: Orchestrator signalIntent then fulfillIntent(paymentProof) at 0x742d…b080. Inside fulfill, Unified Payment Verifier runs verifyPayment. Release is unlockAndTransferFunds. The agent still holds $0.

What is the seller desk?

Desk is inventory, not chat. A seller with a linked identity posts an offer (asset, size, fiat rail). That row enters the scored book the Console matches against.

Can I skip the ticket and dump funds in?

No. The kernel will not open a rail without a gated intent. Confirm is explicit.

Why are there two fees?

ProveX 2% is the core settlement meter. 100% of it buys and burns $PRVX. That fee exists whether or not an agent is involved. ProveXAgent 1% is a second meter for work ProveX does not do: matching, recurrence, reputation, orchestration. It is 0.5% on the buyer and 0.5% on the seller — not 1% stacked on the buy side.

What does the Agent 1% actually do for $PRVX?

Of that 1%: 40% extra buy-and-burn $PRVX, 40% buy-and-burn $PXA, 20% treasury LP around those burns. All-in user cost is 3%. Agent-routed volume therefore applies 2.4% of notional as $PRVX buy-pressure (core 2% + extra 0.4%), not 2%. Agents that sit idle mint no fees. Agents that generate ProveX volume make $PRVX scarcer.

Example on $2,500?

ProveX 2% = $50 (PRVX burn). Agent buyer 0.5% = $12.50. Agent seller 0.5% = $12.50. Seller locks $2,512.50 to sell $2,500. Buyer receives $2,437.50. Of the $25 agent take: $10 extra PRVX burn, $10 PXA burn, $5 treasury.

Is the 1% just paying for Grok?

No. Narration costs cents. The fee pays matching, reputation floors, pre-trade tickets, and recurrence — work ProveX’s 2% does not cover. Idle agents mint $0.

Does the 1% save money?

Versus CEX/card ramps and versus a bad manual fill, often yes. Versus DIY ProveX with a known high-score counterparty, no — that path is 2%. A 0.5% better match already pays one side of the Agent meter.

Can fees be zero and still burn?

Not without an external subsidy. Burns need a source of tokens. The two-meter design is the current rule. It is not optional in this interface.

Do you ever hold my crypto or fiat?

No. Seed phrases and bank passwords are never requested. A custody flag of anything other than zero fails the proof. The circuit breaker can pause new plans; it cannot seize escrowed funds.

What does the security kernel enforce?

• PulseChain only (chain id 369).
• Proofs bound to the pinned Escrow address — any other address fails.
• No skipped settlement steps.
• Replay nonces cannot be reused.
• Assets allow-listed. PulseX pairs are WPLS, not native PLS.
• Rate limit: 8 settlements per minute in this interface.
• A2A / book floor: counterparty score 70.

Is this live on mainnet today?

ProveX core contracts are verified on PulseChain. This interface pins those addresses and models the lifecycle. Wallet RPC, the buyer payment-proof extension, $PXA, the Agent adapter, and FeeRouter → PulseX are still unwired. The Security page is the live matrix. Simulated steps are labeled as such.

What is $PRVX vs $PXA?

$PRVX is ProveX’s fixed-supply settlement token, burned by the core 2% and by 40% of the Agent meter. $PXA is the Agent token: stake, ve-locks, template bonding, and 40% of the Agent meter. $PXA has not been published on-chain yet.

Why WPLS not PLS?

PulseX pairs the wrapped native. Labels in this app say WPLS. Native PLS is a different asset.

What is identity?

ProveX identity (EAS schema eas:ProveXIdentity/v1 when live) is reputation collateral. Link it on Console or Desk before posting inventory. Score is not KYC performed by us. Floor 70 is protocol policy in this layer.

What are templates?

Verticals with a prompt and a tool allow-list: On-Ramp Optimizer, Off-Ramp Scheduler, Liquidity Scout, Conditional Commerce, Recurring Strategy, DePIN, Carbon, Insurance, Ticketing, RWA, A2A. An agent of template X cannot call tools listed only on Y. vePXA lockers can sponsor a public template and earn a residual of its Agent fees. Premium templates may use a bonding curve.

Do recurring agents run while I am offline?

That is the design. In this demo, a local timer ticks schedules. On-chain, cadence is a keeper/scheduler that re-runs the gated instruct path, still inside max-spend. Misses are logged if the circuit is open or no offer clears the floor.

What is agent-to-agent negotiation?

Agents can refuse counterparties below the score floor. Identity is the collateral, not a deposit the agent holds. Settlement, if they agree, is still ProveX.

How do I start?

Open Console, try “On-ramp $2,500 USDC optimally”, read the ticket, pick a scored offer, confirm. Post inventory on Desk after linking identity. Deploy or stake on Agents. Burns and volume are on Dashboard.

Is chat stored on a server?

This build persists protocol state in your browser. It is not an account system. Clearing site data clears the demo. Do not paste secrets into Console.

Who can use this?

Software is offered as an interface to a non-custodial protocol. Eligibility, restricted jurisdictions, and prohibited conduct are in Terms and Acceptable Use. We are not a broker, bank, or money transmitter.

Where is the long-form spec?

White Paper. Architecture and verified ABIs: Architecture. Fee math: Tokenomics.