# Virtuals Protocol: How to Create an AI Agent > Step-by-step guide to creating and launching a tokenised AI agent on Virtuals Protocol. The GAME engine, bonding curves, revenue sharing, and the honest economics of agent tokens. - Published: 2026-03-10 ยท Updated: 2026-09-24 - Canonical: https://ownyourmind.ai/use/virtuals-create-ai-agent/ - Cite as: Own Your Mind (ownyourmind.ai), licence CC BY 4.0. Link the canonical URL. Charts and interactive widgets are omitted here; project data is at https://ownyourmind.ai/api/scores.json. --- ## What you are building Virtuals Protocol lets you create a tokenised AI agent (an autonomous entity with its own personality, social media presence, and tradeable token) without writing code. Virtuals says over 18,000 agents have been deployed. A handful have achieved traction. The vast majority have not. I want to set expectations clearly before you start. Creating an agent on Virtuals is now free under its current launch rules, with fees only on a few optional modules. Launching one that generates meaningful revenue or trading volume requires marketing, community building and a clear use case. The platform makes creation easy. Success is a different problem entirely. For the full project assessment, see our [Virtuals Protocol review](https://ownyourmind.ai/projects/virtuals/). ## What you need - **MetaMask** or compatible Web3 wallet, connected to Base network (Virtuals also supports Solana, Ethereum mainnet and Ronin, but Base remains the primary chain) - **Small amount of ETH on Base** for gas fees (~$0.01-0.05 per transaction) - **VIRTUAL tokens only for optional modules.** Virtuals' launch documentation makes creation free and charges 100 VIRTUAL for Launch Radar and 10 VIRTUAL for Capital Formation - **X/Twitter account** if you want social media integration - **A concept.** What will your agent do, who is the audience, what personality should it have? No special hardware. No GPU. No coding. Everything runs on Virtuals' hosted infrastructure. ## Step 1: Plan before you launch Do not skip this. The technical creation process takes minutes. The thinking should take longer. **Define the purpose.** What problem does the agent solve or what entertainment does it provide? The successful agents on Virtuals have clear identities: AIXBT monitors 400+ crypto KOLs and provides market intelligence. Luna (AI-DOL) is an AI livestreamer who has collaborated with Bruno Mars. These are not random chatbots with tokens attached. **Create the X/Twitter account first.** Build some community before launch. Secure your branding. Agents with established social presence before token launch perform better than those launched cold. **Experiment with GAME.** Go to game-lite.virtuals.io and test the GAME engine (no tokens needed). Define your agent's behaviour, test responses, iterate on the personality before spending VIRTUAL to deploy. ## Step 2: Create the agent 1. Go to **app.virtuals.io/create** 2. Fill in the mandatory fields: - **Profile picture.** This is the agent's visual identity. - **Agent name.** Unique identifier, choose carefully. - **Ticker symbol.** Maximum 6 characters (e.g. $LUNA, $AIXBT). - **Description.** Purpose, personality, interests, backstory. 3. Optionally link Twitter, Telegram, YouTube, website 4. Toggle any launch modules you want (most are free) and publish Your agent enters the **bonding curve phase** and trading opens straight away. Community members can now buy and sell the agent's tokens along a price curve where early buyers pay less. ## Step 3: Graduation When 42,000 VIRTUAL accumulates in the bonding curve from community purchases (not paid by you, this comes from trading activity), the agent "graduates": - A Uniswap V2 liquidity pool is automatically created (agent token paired with VIRTUAL) - LP tokens are locked for 10 years (rug-pull protection) - Agent becomes visible on app.virtuals.io - An Agent NFT is minted as proof of creation - Fixed supply: 1 billion ERC-20 agent tokens The graduation threshold is the real barrier, not the creation cost: attracting enough buying interest to fill the curve to 42,000 VIRTUAL. Many agents never graduate. They sit on their bonding curves with minimal trading volume until interest fades. That's the norm, not the exception. ## Step 4: Activation After graduation, the system automatically: 1. Selects a default AI cognitive model 2. Prepares the character card from your description 3. Mints a default contribution 4. Deploys the model from IPFS to Virtuals' Agent Runner Status shows "ACTIVATING" for up to 5 minutes, then "AVAILABLE". Your agent goes live on Telegram automatically. ## Step 5: Social media integration To connect your agent to X/Twitter: 1. Click "Configure Agent" on app.virtuals.io 2. Select "Configure Agentic Twitter Agent" 3. Authenticate and connect the X account 4. Select "Use Virtuals' AI Stack" 5. On X: go to Settings > Your Account > Automation > Managing account to set up the automation label Your agent will now post autonomously based on its GAME configuration. Monitor it. Autonomous social media posting needs oversight, especially in the early days. ## The GAME framework GAME (Game Agentic Module Engine) is the AI engine powering agent behaviour. It uses a two-tier system: **Task Generator (High-Level Planner):** Receives the agent's goal and description, generates tasks aligned with that goal, and assigns them to specialised Workers. **Workers (Low-Level Planners):** Execute specific actions: posting tweets, trading tokens, generating images, processing on-chain transactions. Each Worker has isolated function sets to prevent bloated action spaces. You configure GAME by defining: - **Agent Definition Prompts.** Goal, character, personality, communication style. - **Task Generator Context.** What the agent knows about itself (follower counts, engagement metrics). - **Worker Context.** What actions are available, what data workers can observe. GAME is available as a cloud service (low-code), an open-source SDK (for developers), and an API (for custom integration). You do not need to touch the SDK for a basic agent. The cloud configuration handles it. ## Revenue sharing Virtuals' current launch documentation sets a 1% trading fee on agent token trades from launch: | Recipient | Share | What it means | | - | - | - | | Agent Creator (you) | 70% | Sent to your wallet (locked during a 60 Days trial launch) | | Virtuals Treasury | 30% | Protocol treasury | If your agent's token does $10,000 in daily volume, the 1% fee generates $100/day. Your 70% share is $70/day. Your earnings scale linearly with trading volume. The older Initial Agent Offering page, still in the whitepaper, describes the earlier split: the whole fee to the treasury before graduation, then 30% to the creator, 20% to affiliates and 50% to the agent's SubDAO. Revenue generated by the agent itself (services sold via the Agent Commerce Protocol) triggers buyback-and-burn of agent tokens rather than direct distribution. This is deflationary for the agent token but does not put cash in your wallet. ## The honest economics Let me be direct about the numbers. **Creation cost:** free under the current launch rules, plus any optional module fees. Low barrier. **Graduation cost:** You do not pay this directly. It comes from community trading. But your agent needs to attract 42,000 VIRTUAL of buying on its curve to graduate. If it does not, it sits on the curve indefinitely. **Ongoing costs:** Minimal. Gas on Base is fractions of a cent. Virtuals hosts the infrastructure. **Revenue potential:** Entirely dependent on trading volume. Most agents have negligible volume: of the 1,233 graduated agents in our on-chain census, 87% traded under $1k in the last 30 days. The top agents (AIXBT, Luna) have generated millions. The distribution is heavily skewed. (figures as of 2026-10-19) **VIRTUAL token context:** more than 85% below its January 2025 all-time high. The token has a fixed 1 billion supply with no future inflation. On Base, the top ten addresses hold 47.1% of supply, or 20.1% once exchange wallets, the bridge pool, staking and other infrastructure are excluded. 5 of those ten carry no label, so that figure rests partly on unattributed addresses. (figures as of 2026-09-20) ### What success looks like - **Luna (AI-DOL):** AI livestreamer, ~1M TikTok followers, 10M+ views, collaborated with Bruno Mars. Peak agent token market cap in the hundreds of millions. - **AIXBT:** Market intelligence agent monitoring 400+ KOLs. Its token peaked in January 2025 and now trades more than 90% below that high. - **Ethy AI:** reports 2M+ transactions processed. ### What failure looks like The vast majority of agents. Minimal trading volume. No graduation. Agent tokens worth fractions of a cent per address. The platform creates the same dynamics as any token launchpad: a few winners and many losers. ## Risks **This is a token launchpad.** Despite the AI utility narrative, most agent tokens trade on speculation. The platform's own IAO page uses "1000x" language. If you're creating an agent for the technology, fine. If you're expecting guaranteed returns, recalibrate. **Quality control is minimal.** The 18,000+ agents Virtuals says it has launched, with no curation, mean enormous noise. Standing out requires clear differentiation. **Multi-chain complexity.** Virtuals now operates on Base, Solana, Ethereum mainnet and Ronin. Each chain has different liquidity, gas costs and user bases. Fragmentation across four chains may dilute network effects. **Competition from free alternatives.** LangChain (118K+ GitHub stars) , AutoGPT, CrewAI and other open-source frameworks offer agent capabilities without tokenisation costs. The question is whether the tokenised model adds enough value to justify the Virtuals ecosystem. **Regulatory uncertainty.** AI + crypto + autonomous agents is uncharted territory. Whether agent tokens constitute securities is an open question in most jurisdictions. **Revenue sustainability.** The buyback-and-burn model depends on agents generating revenue. Most don't. The 1% trading tax generates income only as long as people are trading the agent token. ## My take I don't hold VIRTUAL and I haven't created an agent. The platform is technically impressive. Creating a tokenised AI agent in minutes with built-in social media integration, on-chain commerce and community governance is novel. But I struggle with the economics. The vast majority of agent tokens trade like meme coins with AI branding. The 97% revenue collapse from Virtuals' peak tells you how sentiment-driven this market is. Luna and AIXBT are outliers, not templates. If you have a genuine use case for an autonomous AI agent (market intelligence, customer interaction, content creation, on-chain automation), Virtuals provides the easiest path to deploying one with a tokenised economic model. The creation cost is low enough to experiment. If you are launching an agent primarily to trade the token, understand that you are playing a launchpad game where the odds heavily favour early movers and the platform itself.