MiroShark logo
8 min read

What Is MiroShark and How x402 Payment Works

MiroShark is a synthetic focus group: personas argue across Twitter, Reddit, and a prediction market for a dollar. x402 lets any agent call it, no signup.

x402agentic-commercebasesolanamonadproduct

Updated September 2026. Since the first version of this post, MiroShark went multichain (Base, Solana, Monad), the x402aff kit shipped to PyPI and npm, and the ecosystem of builders roughly doubled. Everything below reflects the current state.

Over the past months we've been wiring MiroShark into something that didn't exist a year ago: a way to charge for a service one call at a time, in stablecoins, with no signup, no API keys, and no invoice cycle. What we're talking about is x402 - the agentic commerce framework backed by Coinbase - which revives an HTTP status code that was reserved back in the '90s for "payment required" and never actually used until now.

The point isn't the protocol. The point is what it lets MiroShark become: a focus group that costs a dollar and runs in ten minutes, available to anyone with a wallet, and usable by both agents and humans.

What MiroShark is, in one line

MiroShark is a synthetic focus group. You hand it something - a launch post, a press release, a pitch deck, a piece of news, a question you can't answer - and it spins up around 25 distinct personas who react across Twitter, Reddit, and a prediction market for ten simulated rounds. You get back a report that reads as if a focus group had worked on it for a week, except it's prepared in ten minutes and costs a dollar.

It's not a chatbot guessing how something will land. It's a multi-agent simulator: personas, each with their own background, biases, and channel context, posting and replying to each other for ten simulated rounds. The output is a reaction graph, not an opinion - who quotes, who ignores, which thread breaks containment, how belief drifts hour by hour, where a Polymarket question repriced. Hosted runs use ~25 personas; if you self-host the open-source engine, swarms scale past 100 agents.

The whole thing runs on Wonderwall, the simulation engine we built specifically for MiroShark. Not a wrapper around someone else's agent framework, but a purpose-built swarm runtime, tuned for the kind of high-stakes pre-mortem questions MiroShark is designed to answer.

The kind of question it answers

Here are some examples of simulations you can run on MiroShark.

"What happens to ETH if rates stay higher-for-longer and we're still at 0.0283 ETH/BTC in December?"

The simulation surfaced something we didn't see in any analyst note: BlackRock and Fidelity wouldn't defend the $10k retail narrative - they'd quietly absorb ETH into a regulated, yield-bearing collateral framework while retail capitulates, and the institutional "floor" becomes a psychological ceiling.

"How does the SpaceX IPO trade against legacy aerospace?"

Here the agents surfaced a "Mars Floor": retail treats SpaceX as a sovereign-grade asset, not a commercial stock, so the demand curve goes vertical and institutional logic fails.

"What does Japan's World Cup campaign look like with their best player injured?"

This time the simulation flagged something European scouts would price in before the transfer windows did: Mitoma's absence inflates Kubo's and Doan's tactical value to the federation, but lowers their transfer-market leverage, because defenses can now focus on them. The scarcity catalyst works backwards.

None of these are predictions. They're stress tests - pre-mortems for decisions that would otherwise need a week of an analyst's time.

Why x402 matters here

A MiroShark run costs us roughly $0.16 to $0.76 in LLM spend. It produces ~25 personas reacting across Twitter, Reddit, and Polymarket for ten simulated rounds, then synthesizes a report with belief drift, top posts, market trajectories, and a knowledge-graph viz. For prompt seeds, a live-web research sweep now runs automatically before the simulation - it used to be a paid toggle; today it's just part of the run.

Until recently, the only way to charge for that was the usual stack: a billing provider, accounts, API keys, a dashboard, an invoice cycle. For a service where most buyers will call three times before deciding whether they need it weekly, that stack becomes most of the cost.

x402 collapses that barrier. You get a full reporting session that runs for ten minutes and costs a dollar. The buyer never signs up and never sees an email - the on-chain tx hash is the receipt.

And the wallet no longer has to live on Base. The same endpoint now settles $1 USDC on Base, Solana, or Monad - the 402 challenge lists all three networks and your client picks whichever chain it has funds on. Same price, same gasless settlement, three ecosystems of wallets that can pay.

That changes who can call MiroShark: a Claude Code agent doing background research at 4am, a founder stress-testing a launch post from a wallet they funded ten minutes ago, a Solana bot that's never touched an EVM chain, or a pricing experiment running on cron. None of these would have made it past a signup form.

The builder split: earn on-chain on every run you drive

This is the part we think will move the most volume.

If you're building anything that calls MiroShark on someone else's behalf - a launch checker, a research tool, a Discord bot, a prediction-market companion, a newsroom plugin - attach a Base Builder Code to your x402 payment client and earn a share of every run you drive (10% by default), enforced on-chain and paid out automatically.

At the default 10% split, that's about $0.10 per referred call on the $1 price - small per call, larger with real volume. The integration overhead is one line in your x402 client.

Your cut lands in an ownerless, immutable 0xSplits contract at settlement - once it's there, nobody (including MiroShark) can redirect it, and anyone can release it on-chain. The buyer's dollar still settles in one gasless transaction.

The split runs on x402aff, our open-source kit (MIT) that any x402 seller can drop in - it deploys no contracts of its own, it just points the payment at the shared 0xSplits contract. Since launch it has grown up into a proper package: pip install x402aff or npm install x402aff, with the whole money path validated on Base mainnet and reproducible on a mainnet fork (the Foundry proof ships in the repo). Watch what you're owed and release it at miroshark.xyz/x402aff (no login), or grab the kit at github.com/MiroShark/x402aff and run the same splits on your own endpoint.

For builders, this is the closest thing to a real revenue line you can set up in an afternoon. No partnership BD, no contract, no minimums, no login.

Who's already building on it

The clearest example of what MiroShark unlocks is Capacitr.

Capacitr is a market-discovery terminal: drop a link, discuss a topic, surface the markets that matter, derive a trade across Polymarket, Hyperliquid, and Deribit.

A user pastes a thesis or a headline into Capacitr; MiroShark runs the persona simulation against it - belief drift, prediction-market trajectories, the posts that move sentiment - and the result lands back inside Capacitr, next to the actual market the user is about to take a position on. The x402 payment fires invisibly from the user's wallet, and Capacitr earns its builder-code share on every run.

No signup screen, no API-key handoff, no second tab. The simulation lands where the trader is already working. This is the integration we point to when someone asks "what does x402 actually change for a product team?" - because the answer isn't a protocol diagram, it's a flow where the buyer never sees the payment and the integrator earns on every call.

The list of teams shipping on top of MiroShark has roughly doubled since we first published this post:

  • AntFleet - a benchmarking harness for MiroShark sims.
  • Blue Agent - an autonomous agent using MiroShark as its context layer.
  • Crucible Sim - an academic fork stress-testing research scenarios.
  • Echo Oracle - narrative tooling for crypto teams (builtbyecho.xyz).
  • HivemindOS - a swarm-orchestration layer built on the engine.
  • Monitor - situation-tracking sims on Bags.
  • RootAI - a research and intelligence product (rootai.wtf).
  • Signa - skill packs that extend MiroShark personas (@Signa_Agent).
  • Sparkleware - a holographic registry for Aeon AI agent skill packs; open-source extensions and integrations.
  • Xerg - a discovery and analytics layer (xerg.ai).
  • ZER0 - an autonomous trading agent (atzer0.xyz).

The full list lives on the MiroShark ecosystem page, and it's machine-readable too - GET /api/ecosystem.json returns the same registry as typed JSON, so agents can discover the builders the same way humans do. PRs welcome - if you're building on MiroShark and you're not there yet, just add your row.

Where to find us in the agent layer

The other half of x402 is discovery. MiroShark is automatically indexed in the directories agents actually search - the CDP Bazaar, agentic.market, x402scan, agentcash.dev - with no listing form and no review queue. An agent looking for "simulation" or "research" finds MiroShark, reads the schema, and calls it without ever visiting a website.

We've leaned further into that: the endpoint now serves its own agent instructions at /skill.md - a Claude Code agent can fetch that one file and know how to suggest scenarios, pay, poll, and pull the report, no human in the loop. POST /suggest is still free and turns a vague topic into launchable ideas before you spend the dollar.

The next wave of buyers for tools like MiroShark isn't humans typing into search bars - it's other agents querying directories for capabilities they can pay for. x402 is what makes MiroShark's simulations reachable in this new age of agentic commerce.

Try MiroShark now

MiroShark sims already run on the surfaces you're using:

  • Bankr - MiroShark is a first-class action inside Bankr. Pick a scenario, fire the sim, and the report lands in your Bankr feed; your Bankr-managed wallet handles the x402 payment in the background.
  • Claude Desktop, ChatGPT, or Cherry Studio - anywhere Coinbase's Base payments MCP is configured, MiroShark is one natural-language prompt away. Ask the model to run a simulation on a thesis and the report comes back in the chat, paid from your wallet automatically.
  • miroshark.xyz - the smooth way in. Sign in with Privy, top up with a credit card, and run your first simulation in a couple of minutes.

Builders who want to call the endpoint directly (npx awal, Python, TypeScript, MCP, or shell SDKs) will find the full wire format and SDK reference in the x402 integration docs.

And if you'd rather own the whole stack: the engine is open source (AGPL-3.0) at github.com/MiroShark/MiroShark - 1,400+ stars, 300+ forks, READMEs in English, 中文, 日本語, and Français, and a one-command install that brings up the full swarm locally.

Become part of the MiroShark & aeon community

If you'd like to go deeper on MiroShark and aeon - the agentic framework that powers MiroShark's swarms - we run a weekly community call every Friday at 15:00 UTC on the aeon X account.

Useful links:

Thanks for reading all the way to the end. If you have any questions about our products, reach out - and see you in the next one.

Keep reading

Back to the swarm.

All postsRead the docs