> For the complete documentation index, see [llms.txt](https://docs.velvet.capital/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.velvet.capital/velvet-socialfai-ecosystem/ai-framework.md). # AI Framework Velvet’s AI framework is the intelligence layer of the platform. It has two core components: * **Velvet Unicorn (VU)** — a multi-agent AI framework that powers research, market intelligence, and user-directed execution across the platform. The framework contains the skills, tools, and plugins to deliver fast, real-time information about current crypto-market conditions. * **Velvet-1** — Velvet’s crypto-native AI model, trained on crypto and onchain data and integrated with tools from Velvet Unicorn and beyond, to power the framework’s intelligence — spanning onchain analysis, raw onchain data graph analysis, and CEX market data. All actions are initiated and controlled by the user. ### Velvet Unicorn (VU) Velvet Unicorn is the AI assistant at the heart of the Velvet DeFAI ecosystem. It combines onchain data, market intelligence, and execution capabilities into a unified agent system that analyzes opportunities, generates insights, and executes user-requested actions through natural language. #### Architecture: a multi-agent system Every VU request is routed through a multi-stage agent pipeline. A router first classifies the query and hands it to the right specialist — token analysis, wallet analysis, swap, trending, and dedicated pipelines for Hyperliquid, top-trader data, and the Virtuals ecosystem. For deep analysis, specialist agents run in parallel: a technical analyst for price and volume data, a sentiment agent for social and news coverage, and a fundamentals agent for tokenomics and protocol context. Alongside the large LLM-based reasoning agents, the system uses smaller, custom-built, task-specific models (non-LLM) for precision functions such as statistical market prediction and classification. This hybrid design reduces the hallucination risk that makes general-purpose AI unreliable in financial contexts. The system improves through a structured evaluation process: outputs and execution outcomes are measured against results, and the team uses this feedback — drawn from both onchain and social data — to refine agent routing, prompts, workflows, and model accuracy over time. #### What VU can do * **Research & discovery** — trending-token discovery from a plain-language instruction: filter by chain, category, launch date, volume patterns, and social activity, with key metrics and an AI summary for each result. * **Token analysis** — full multi-agent analysis of any token (by symbol, contract address, or plain-language name; chain auto-detected), combining onchain data, price history, liquidity depth, social sentiment, and top-holder activity into a single structured report. * **Wallet & portfolio analysis** — open-ended questions about any EVM or Solana wallet: performance, PnL history, top positions, trading frequency, risk exposure, and cross-chain activity — useful for due diligence and copy-trading research. * **Market intelligence** — real-time insights from onchain, market, and social data; AI-generated market observations and news digests are surfaced in the Velvet feed. * **Execution** — intent-based commands that go from analysis to action. A natural-language instruction like “swap 0.5 ETH for USDC” is resolved into tokens and the best available quote among integrated sources, returned as a ready-to-sign transaction that executes from the user’s own wallet. Swap execution is supported on Ethereum, Base, Solana, and BNB Chain. * **Platform assistance** — help navigating Velvet, campaigns (staking, Gems), and troubleshooting. VU’s market intelligence spans all chains supported by the platform, with dedicated coverage for Hyperliquid queries. Natural-language swap execution currently covers Ethereum, Base, Solana, and BNB Chain, and Prompt-to-Strategy runs on Base. #### Build with VU: API, MCP, and Prompt-to-Strategy The Velvet Unicorn agent portal opens the intelligence layer to builders and other AI agents: * **API access** — developers and AI agents can call VU’s endpoints directly (token analysis, trending, wallet analysis, swaps, and open-ended chat). Two ways to pay: sign up for an API key with 1,000 free requests, or go accountless with x402, Coinbase’s open payment protocol for AI agents — the API responds with an HTTP 402 payment request, and the agent pays per call in stablecoins directly over HTTP. VU is also registered on the Virtuals Agent Commerce Protocol (ACP), so autonomous agents can discover, hire, and pay it onchain. * **MCP support** — the same capabilities are exposed as native MCP tools for Claude Code, Codex, and any MCP client, plugging DeFi intelligence directly into agent-native environments. * **Prompt-to-Strategy** — turn natural-language ideas into live, running DeFi strategies. An agent builds a structured strategy spec (rebalances, DCA, scheduled sells, and more), resolves tokens to contract addresses, preflights it against the account’s wallet, and deploys it. Velvet injects the spec into an audited execution template and launches it as a live process, with status, an onchain action log, and a stop control — every transaction signed by the account’s own non-custodial wallet. Strategies execute only the rules defined by the user; Velvet does not select, manage, or adjust strategies on a user’s behalf. Explore the agent portal: [vu.velvetdao.xyz/agent](https://vu.velvetdao.xyz/agent/) — including an in-browser playground where signing up creates a non-custodial wallet and a free API key automatically. Read more: [Velvet Unicorn AI Goes Live on x402](https://blog.velvet.capital/p/velvet-unicorn-ai-goes-live-on-x402) ### Velvet-1 Velvet-1 is Velvet’s crypto-native AI model, developed by the Velvet team and trained on crypto and onchain data to serve as the model layer of the Velvet AI framework. * **Purpose-built for crypto.** Velvet-1 is trained on crypto-native data — onchain data, market data, and social data — so it understands tokens, onchain markets, and DeFi context natively. * **Powers the Velvet AI framework.** Velvet-1 works inside the VU agent system — including the VU MCP and Prompt-to-Strategy — as the model layer behind research, market intelligence, and strategy generation. #### Why crypto needs its own model Onchain analysis has remained largely unsolved for AI. Today’s AI models excel at reasoning over static information such as code, documentation, books, and web pages. Unlike text, blockchain data is naturally represented as massive, interconnected graphs: every wallet, transaction, smart contract, protocol, liquidity pool, and token is connected to countless others through relationships that are constantly changing. Those graphs evolve every second as new transactions settle, liquidity moves, positions open and close, and entirely new networks of interactions emerge. To truly understand what’s happening onchain, an AI model can’t simply retrieve information. It needs to reason over evolving graph structures — to understand the full context of onchain activity, market structure, and social dynamics as one interconnected, continuously evolving system. Building a model capable of this kind of reasoning involves large volumes of onchain data and extensive research into new layers, context mechanisms, and training techniques designed specifically for continuously evolving onchain environments. ### Non-custodial by design Across the framework, AI never takes custody of user assets. In the terminal, VU executes from the user’s own connected wallet through the Velvet execution engine; API swap calls return a ready-to-sign transaction that the user’s wallet submits; and Prompt-to-Strategy strategies sign every transaction with the account’s own non-custodial wallet, created during onboarding. > Velvet’s AI features are tools for research and execution — they are not financial advice, and outputs are not recommendations to buy or sell any asset. AI can make mistakes; always do your own research. See our [Terms of Service](https://www.velvet.capital/terms-of-service.html) for eligibility and jurisdictional restrictions.