# MiroShark

> Simulate anything for ~$1 in under 10 minutes. You bring a scenario; MiroShark spawns hundreds of grounded AI agents that post, argue, trade on a prediction market, and change their minds hour by hour - then hands you a report citing the actual posts and trades.

MiroShark is a universal swarm-intelligence engine. Drop in any scenario - a press release, a news headline, a policy draft, an unanswerable question, a historical what-if - and it builds a world of hundreds of LLM-powered agents that react to it over simulated time on a Twitter-like feed, a Reddit-like forum, and a prediction market. You can chat with any agent, inject breaking news mid-run, and fork the timeline. Runs are grounded in a knowledge graph (Neo4j) and driven by open models via OpenRouter. Payment is pay-per-run through the x402 protocol on Base, Solana, or Monad (~$1 per simulation); there is no subscription.

## When to use MiroShark
- Use it to explore counterfactuals and "what-if" scenarios: how a market, community, or public might react to an announcement, policy, product, or event.
- Use it for multi-agent / swarm simulation, narrative forecasting, or red-teaming a message before a launch.
- Launch a run at https://www.miroshark.xyz/sim. The first simulation costs about $1 of LLM credits, paid per-run via x402 - no account tier or subscription required.
- Do not treat outputs as guaranteed real-world predictions, financial advice, or ground-truth facts: they are simulated agent behavior, not certainties.

## Docs
- [How MiroShark works](https://www.miroshark.xyz/docs): the pipeline - document to graph build to persona grounding to a 3x platform loop to a cited report.
- [x402 integration guide](https://www.miroshark.xyz/docs#x402): how pay-per-call x402 payments work and how to integrate them.
- [x402aff dashboard](https://www.miroshark.xyz/x402aff): builders earn 10% of each run they drive, paid on-chain into an ownerless 0xSplits contract on Base; shows what each builder is owed, lets anyone release it, and indexes every split the open-source x402aff kit has routed across all sellers.
- [x402aff](https://www.x402aff.xyz): the open-source (MIT) builder-code affiliation kit the dashboard runs on - add on-chain affiliate splits to any x402 seller; on npm and PyPI as x402aff. [Developer docs](https://www.x402aff.xyz/developers), [llms.txt](https://www.x402aff.xyz/llms.txt).
- [Ecosystem](https://www.miroshark.xyz/ecosystem): projects and partners building with MiroShark.
- [Blog](https://www.miroshark.xyz/blog): writing on MiroShark, agent simulation, and x402.
- [Changelog](https://www.miroshark.xyz/changelog): product updates.

## Run a simulation
- [Launch the run panel](https://www.miroshark.xyz/sim): describe a scenario and start a simulation (pay ~$1 via x402).

## Pricing
- About $1 per simulation, paid per-run in USDC via the x402 protocol on Base, Solana, or Monad. No subscription.

## Developer resources (by name)
- MiroShark API documentation (integration guide): [https://www.miroshark.xyz/docs#x402](https://www.miroshark.xyz/docs#x402).
- MiroShark OpenAPI spec (site API): [https://www.miroshark.xyz/openapi.json](https://www.miroshark.xyz/openapi.json) - typed, function-calling-ready description of /ask, health, ecosystem, and validate-url. Versioned (v1); every response carries an `API-Version` header, and abuse-bearing endpoints return `RateLimit-*` headers.
- MiroShark OpenAPI spec (x402 simulation API): [https://x402.miroshark.xyz/openapi.json](https://x402.miroshark.xyz/openapi.json) - the pay-per-run contract.
- MiroShark authentication: x402 v2 - the payment is the auth (no API key, no OAuth). See [https://www.miroshark.xyz/docs#x402](https://www.miroshark.xyz/docs#x402).
- MiroShark MCP server (Streamable HTTP): POST [https://www.miroshark.xyz/api/mcp](https://www.miroshark.xyz/api/mcp); manifest at [https://www.miroshark.xyz/.well-known/mcp.json](https://www.miroshark.xyz/.well-known/mcp.json). Tools: ask_miroshark, suggest_scenarios, miroshark_overview, miroshark_run, miroshark_resources, get_ecosystem, get_health.
- MiroShark open dataset (Hugging Face, MIT): [https://huggingface.co/datasets/MiroShark/social-prediction-market-sim](https://huggingface.co/datasets/MiroShark/social-prediction-market-sim) - 8,201 agent decisions from 16 simulations, social feed actions and prediction-market trades, with prompts, reasoning, and actions (SFT-ready parquet).

## For agents (machine-readable)
- [/index.md](https://www.miroshark.xyz/index.md): this summary as a markdown document (append .md to the root, or send Accept: text/markdown).
- [/agents.md](https://www.miroshark.xyz/agents.md): the agent guide.
- [OpenAPI spec](https://www.miroshark.xyz/openapi.json): typed description of this site's public endpoints (/ask, health, ecosystem, validate-url) for function calling.
- [x402 API OpenAPI](https://x402.miroshark.xyz/openapi.json): the pay-per-run simulation API contract.
- [MCP server](https://www.miroshark.xyz/api/mcp) ([manifest](https://www.miroshark.xyz/.well-known/mcp.json)): MiroShark as MCP tools over Streamable HTTP.
- [Agent Skills index](https://www.miroshark.xyz/.well-known/agent-skills/index.json): discovery index with a content digest.
- [API catalog](https://www.miroshark.xyz/.well-known/api-catalog): RFC 9727 linkset of the API surfaces.
- [ai-catalog](https://www.miroshark.xyz/.well-known/ai-catalog.json): ARD manifest naming every agentic resource.
- [/ask](https://www.miroshark.xyz/ask): NLWeb endpoint - POST a natural-language question, get ranked schema.org results (add `prefer: streaming=true` for SSE).

## Optional
- [GitHub](https://github.com/MiroShark/MiroShark): source and stars.
- [X / Twitter](https://x.com/miroshark_): updates and contact.