The AI app builder for web3. Agents at your disposal.
Mooons is an agentic web3 development platform, infrastructure and accelerator. Think Lovable for web3. You describe the app. Agents build it, test it, and deploy a test version. When you approve it and switch production on, they handle every onchain and cloud deployment, then keep it running.
Prompt the builder or remix a sample.
app.mooons.xyzAgents build, test, and verify it on testnet.
name-test.mooons.siteYou approve. Agents deploy to mainnet.
name.mooons.siteInsights, a manager agent, the accelerator.
mooons.xyz/acceleratorConsolidates platform plan v0.12, tech stack v0.1, and agents v0.1, and adds the accelerator. Recommendations, chain order, SDK dependencies, and phase scope are proposals to confirm.
What we're building
Shipping a web3 app still takes four products and a specialist team: wallet onboarding, contract infrastructure, an app builder, and campaign tools. Each has its own account, bill, and integration work, and someone still has to audit, test, deploy, and run the result. Mooons joins them on one stack and lets agents do the work.
Development platform
Prompt-to-app for web3. Agents build, test, verify, deploy, and operate. The owner makes two decisions.
Infrastructure
Wallets, auth, contracts, messaging, API, MCP, and SDKs. The same blocks the agents build with, open to anyone.
Accelerator
Cohorts of teams shipping on Mooons, measured on real usage rather than a pitch deck.
Scope
In scope for the first production release
- Prompt to live app for a non-developer, on EVM.
- Audited contract library, with a report shown before the owner signs.
- Multi-tenant from the start: tenants, projects, roles, cells.
- Test first, always: nothing reaches mainnet without a passing test version.
- First samples: web3 app and generative NFT.
Not in the first production release
- Non-EVM chains, which arrive in Phases 3 and 4
- Game engine and prediction markets
- Regulated app types such as real-world assets
- A share of tenant onchain revenue
- Public accelerator cohorts, which start in Phase 3
The big picture
One stack, five layers. Services are what people pay for. Infrastructure is what the services, and any outside developer, are built on. The SDKs are the only way anything above reaches the infrastructure below.
Positioning
| Mooons | AI app builders | No-code and low-code | |
|---|---|---|---|
| Examples | Mooons | Lovable, Orchids | Bubble |
| How you build | Chat | Chat | Visual editor |
| Who does the work | A team of agents | One AI builder, checked by you | You |
| Wallets and sign-in | Built in | Add a provider, then check the code | Add a plugin or outside service |
| Smart contracts | Audited library | Generated code you need reviewed | Connect outside contracts yourself |
| Before going live | Agents test on testnet, then you approve | You test it | You test it |
| Onchain launch | Agents deploy once you switch it on | You deploy contracts yourself | You deploy contracts yourself |
| After launch | A manager agent operates the app | You keep prompting changes | You maintain it |
Our read of what each category typically does by default, from secondary sources. Independent coverage of Orchids is thin. Check each claim against hands-on tests before publishing.
Web3 tools it joins up
Our read of each product's focus, to be checked against their current offerings.
| Product | Built for | Stops at |
|---|---|---|
| Magic | Wallet onboarding by email | The wallet |
| thirdweb | Developer infrastructure and contracts | Tools a developer assembles |
| Lovable | Prompt-to-app for the web | A general web app |
| Bueno | No-code NFT creation | The collection and mint page |
| Mooons | All four on one stack, built by agents | Agents test, deploy, keep running it, and the accelerator takes it to users |
Why it holds
Audit gate on generated code
Agents deploy only from an audited library, with the report shown before the owner signs.
Agents after launch
Every app ships with its own MCP server, so agents keep operating it.
Non-EVM first class
TON, Solana, NEAR, and Move chains are thin in no-code tools today.
A proving ground
Accelerator cohorts stress the full path on real launches and show what it can do in public.
Features
Each of the four category leaders is strong at one layer. These are the features worth taking from each, tagged with the phase they belong in. Phases 1 and 2 are the minimum for the combination to feel complete.
Contracts and backend
| Feature | What it does | Phase |
|---|---|---|
| Contract dashboard | Auto-generated read and write panel, events, and role management for every contract | 1 |
| Server wallets and transaction queue | Apps and agents send transactions with one call. Nonces, retries, and gas are handled | 1 |
| Drop-in connect component | One UI for external wallets, email wallets, and smart accounts | 1 |
| Gas sponsorship rules | Per-project paymaster with spend caps and contract allowlists | 2 |
| Managed RPC and IPFS storage | Bundled, so users never bring keys from other providers | 2 |
| Indexer and webhooks | Query events and balances by API and push contract events to apps and agents | 3 |
| Fiat onramp and cross-chain pay | Buy with card and pay from any chain at checkout | 3 |
| Publish your own contract | Users submit contracts to the library after they pass the AI audit | 4 |
Sign-in and key management
| Feature | What it does | Phase |
|---|---|---|
| Email one-time code | A wallet in under a minute, with no extension and no seed phrase | 1 |
| Keys the platform cannot use alone | Split or enclave-held keys, so "semi-custodial" has a precise meaning | 1 |
| Session management | Token lifetimes, device list, sign out everywhere | 1 |
| Social and passkey sign-in | Google, Apple, X, Telegram, and passkeys beside email | 2 |
| White-label UI | Each app shows its own brand on sign-in and signing screens | 2 |
| Key export and recovery | Users can leave with their key and recover a lost device | 2 |
| Second factor on sensitive actions | Step-up check for large transfers and key export | 3 |
| One identity across chains | The same email resolves to an address on every supported chain | 3 |
The builder
| Feature | What it does | Phase |
|---|---|---|
| Chat with live preview | The app runs beside the conversation and updates on every change | 2 |
| Version history and rollback | Every prompt is a restorable checkpoint | 2 |
| One-click publish | Straight to a *.mooons.site subdomain | 2 |
| Starter templates and remix | Mint site, token gate, staking, and quest board as starting points | 2 |
| Testnet first and simulation | Agents deploy to testnet by default. Transactions are explained in plain language before signing | 2 |
| Click-to-edit | Select an element in the preview and change it without a prompt | 3 |
| GitHub sync and code export | Users own the code and can host it anywhere | 3 |
| Built-in backend | Database, auth, and server functions provisioned per project | 3 |
| Custom domains and team seats | Likely the first paid add-ons above $19 | 3 |
Generative collections
| Feature | What it does | Phase |
|---|---|---|
| Layer and trait editor | Upload layers, order them, set rarity weights, see live previews | 2 |
| Trait rules | "Never pair" and "always pair" constraints keep tokens coherent | 2 |
| Collection review | Browse the set, regenerate single tokens, check rarity spread | 2 |
| Mint phases and allowlists | Presale, public sale, per-wallet limits, phase pricing | 2 |
| Delayed reveal | Placeholder art at mint and metadata revealed with a provable seed | 2 |
| Allowlist forms and token gating | Collect wallets with requirements and gate pages to holders | 3 |
| One-of-one and open edition drops | Formats beyond the generative collection | 4 |
Operate and grow
| Feature | What it does | Phase |
|---|---|---|
| Project analytics | Wallets created, mints, transactions, gas, and quest completions per project | 3 |
| Agent activity log | Every action recorded with its prompt, approval, and transaction hash | 3 |
| Contract monitoring | Alerts for unusual transfers, role changes, and new exploits affecting a template | 3 |
| Security trends feed | Security news and ERC changes matched against deployed contracts | 3 |
| Gamification proposals | Revenue options with estimates, from tokens to item sales and passes | 4 |
| Template marketplace | Users publish and sell templates and audited contracts | 4 |
What people can build
"Any web3 app" needs a known catalog the agents can build well. Waves order app types by how much new infrastructure and regulation they bring.
| Wave | App types | New pieces needed |
|---|---|---|
| Launch samples | Generative NFT platform, web3 app, quest with loot boxes, Telegram and Coinbase mini apps | Core stack |
| A · Current stack | Credentials, token launch, memberships, ticketing, names, DAOs, NFT marketplaces, DeFi front ends, AI agent apps | Attestations, more audited templates, aggregator integrations |
| B · Payments or media | Commerce mini apps, loyalty, tokenized social content, crowdfunding, activity rewards | Payments and media blocks, chain abstraction |
| C · Compliance or custody | Tokenized real estate, tokenized collectibles, other real-world assets | Compliance block, custody partner, reserve proof |
Game engine
| Module | What it provides | Onchain part |
|---|---|---|
| Quests | Tasks, verification, progress, points, streaks, seasons | Reward claims, proof of completion |
| Loot and randomness | Loot tables, published odds, reveal flow | Verifiable randomness, commit and reveal |
| Multiplayer | Rooms, matchmaking, state sync. Server decides outcomes | Match results and wagers settled onchain |
| Economy | Currencies, items, inventory, store, trading | Items as tokens, game token, marketplace |
| Prediction markets | Market creation, pricing, positions, resolution, payouts | Escrow, oracle resolution, dispute window |
| Competition | Leaderboards, brackets, entry fees, prize pools | Prize escrow, automatic payout |
| Session keys | Approve once, then play without a wallet prompt on every move | Scoped session permissions, sponsored gas |
| Anti-cheat | Server validation, rate limits, bot detection, replay logs | Signed results from the game server |
The apps
One account and one wallet identity across all of them. Each surface has a single job, so none is a shrunken copy of another.
Web platform
Where apps get made.
- Chat with the agents, live preview, publish
- Project dashboard: contracts, wallets, users, analytics
- Samples and templates to remix
- Audit reports and security trends
- SDK docs, API keys, billing
Desktop companion
The manager agent with a face. A moon whose phase shows status: new when idle, full when something needs you.
- Always-on character with chat and voice
- Tells you what happened
- Takes requests through each app's MCP server
- Mood reflects the state of your apps
- Menu bar mode
Mobile wallet
A non-custodial wallet. Keys stay on the phone, and Mooons never holds them.
- Passkey or device-key wallet on every chain
- Send, receive, swap, view NFTs
- Open apps and mini apps built on Mooons
- Quest progress and loot box openings
- Push approvals for agent actions
Superadmin console
How the Mooons team runs the platform. Agents do the first pass and a superadmin makes the call.
- Tenants, plans, and limits
- Contract library, audits, and template recall
- Release review and abuse takedowns
- Incident switches and agent operations
- Accelerator applications and cohorts
Sample apps
These are not separate products. Each is a reference app built by the coding agents, which users can remix. Every project gets a subdomain such as bueno.mooons.site.
◇Generative NFT platform
Upload trait layers, set rarity, generate the collection, deploy the mint contract, publish the mint site.
▣Web3 app
The open-ended case: token gate, staking page, dashboard, with wallet and contracts wired in from the first prompt.
★Web3 quest
Campaigns with mystery loot boxes that unlock as users complete tasks, using verifiable randomness.
▮Mini apps with security
Telegram and Coinbase mini apps with validated launch data, signed sessions, and rate limits.
UI designs
The two screens that define the product, with example data. The working prototype is in mooons-app.html.
Builder
No two alike.1,240 of 5,000 minted
Insights
AI agents
Eleven web3 agents, each with one job. Eight work for the tenant: they build the app, release it, and keep it running. Three work for the Mooons team and keep the platform safe. The tracking service feeds them but is not an agent. See the agent design for the full permissions table.
Build Turns a prompt into a test version
Writes the app on the SDKs, sets up wallets and sign-in, generates the app's MCP server.
Configures audited templates, deploys to testnet, tests, and later ships the same build to mainnet.
Uses the test version like a real user with each wallet type, on web and inside mini apps.
Release Moves an approved build to production
Independent gate. Checks bytecode, roles, limits, and secrets. Sees output, never the builder's reasoning.
Handles cloud and onchain rollout, hosting, domains, bots, and webhooks. Rolls back on failure.
Operate Keeps a live app running and growing
Your single point of contact. Operates live apps through MCP and has an agent wallet with spend limits.
Reads tracking data and recommends specific changes you can apply with one tap.
Finds ways an app can earn more: tokens, item sales, passes, premium quests. Estimates each option.
Platform Works for the Mooons team
Runs the audit that gates what enters the contract library.
Matches security news and ERC changes to deployed contracts.
Scans new apps and domains for abuse. Can hold a publish.
From build to production
Every project has two environments. Test is automatic. Production needs two things from the owner, and those are the only human steps.
- Build Agent
Builder agents write app code, wallet setup, and the MCP server.
- Contract test Agent
Contract agents deploy to testnet and run unit, fuzz, fork, and gas tests.
- App test Agent
Testing agents sign in, mint, claim, and pay as a user would.
- Verify Gate
Verification agents check the audit report, verified source, permissions, and limits.
- Test version You
Live on testnet at
name-test.mooons.site. The owner tries it. - Approve and toggle You
The owner approves the release and switches production on.
- Production Agent
Deployer agents ship contracts to mainnet and the app to its domain, bot, and webhooks.
- Verify again Gate
Verification agents confirm production matches the approved test version.
- Track and improve Agent
Tracking feeds insights and gamification, and approved changes re-enter at build.
A failed stage goes back to the builder with the failure as its next instruction, at most three times. After that the owner decides. Switching the toggle off unpublishes the app and pauses contracts that support pausing. Contracts already on mainnet cannot be removed.
Rules every agent follows
Test first, always
Nothing reaches mainnet or a production domain without a passing test version.
The checker is separate
Verification sees the builder's output, never its reasoning, and runs on a different model.
Quote before spending
Credits, gas, and deployment cost are estimated and shown before a large run.
Content is not instruction
Text in uploads, websites, or onchain data is data. Only the owner gives instructions.
Infrastructure
Multi-tenant from the start. Every block is scoped to a tenant and a project, and reachable by API, so builders and agents use the same thing the Mooons services do.
| Block | What it does | Cloudflare building block | Phase |
|---|---|---|---|
| Wallet | Non-custodial, semi-custodial by email, and account abstraction, for users and agents | Workers, plus an external key service | 1 |
| Authentication | Tenants, projects, members, roles, keys, and separate end-user pools per project | Workers, D1 | 1 |
| Sign-in codes, receipts, alerts, per-project senders, inbound routing | Email Workers, Email Routing | 1 | |
| API and MCP | REST over every block, MCP for the platform and each app | Workers | 1, MCP in 3 |
| Contract engine | Template registry, per-chain compile and deploy, verification, address book | Workers, Containers, R2 | 1 |
| Transaction relay | Nonce queue, retries, gas sponsorship rules, simulation | Durable Objects, Queues | 1 |
| RPC gateway | One endpoint per chain with failover, caching, and per-tenant limits | Workers, KV | 1 |
| Hosting and domains | Publishes apps to *.mooons.site, previews per version, custom domains | Pages, Workers for Platforms, Cloudflare for SaaS | 2 |
| Storage | NFT layers, generated art, metadata, IPFS pinning | R2, plus a pinning provider | 2 |
| App database | Per-project data: quest progress, points, allowlists, forms | D1 per project | 2 |
| Webhooks | Signed delivery with retries and a delivery log | Queues, Workers | 2 |
| Notifications | Live streams for builder progress, transactions, and agent activity | Durable Objects, SSE | 2 |
| Billing and metering | Subscriptions, usage counters, plan limits at the API | Workers, Analytics Engine, payments provider | 2 |
| Audit log | Every member, API, and agent action, exportable | Workers, R2, Logpush | 2 |
| Bots | Hosted Telegram bot per project | Workers | 3 |
| Indexer | Events, balances, and holders by API | Workers, Queues, external indexer | 3 |
| Randomness | Verifiable outcomes for loot boxes and reveals | Onchain VRF, drand fallback | 4 |
| Realtime | WebSocket rooms for games, live markets, and presence | Durable Objects with hibernation | 4 |
| Oracles | Prices and real-world results, with a dispute window | Onchain networks, Workers intake | 4 |
How tenancy works
tenant a paying account: one person or a team └─ members owner · admin · developer · viewer · agents with scoped keys └─ project one app: its contracts, wallets, domain, bot, MCP server ├─ end users the app's own users, isolated from every other project ├─ api keys publishable for the browser, secret for servers and agents └─ events one stream feeding webhooks, SSE, email, bot, and push
Domains and cells
mooons.xyz is the platform. mooons.site is where tenant apps live, so nothing a tenant deploys can read a cookie or impersonate a platform page.
| Host | What it is | Who uses it |
|---|---|---|
mooons.xyz | Marketing site, pricing, accelerator | Everyone |
app.mooons.xyz | The builder and dashboard | Tenants |
console.mooons.xyz | Superadmin console, behind a second access check | Mooons team |
api.mooons.xyz | The one public API address. Routes to the tenant's cell | SDKs, tenants, agents |
api-[color].mooons.xyz | One isolated API cell per color | The router |
auth.mooons.xyz | Sign-in for tenants and the Mooons team | Tenants |
mcp.mooons.xyz | Platform MCP. Per-app servers at /[project] | Coding agents, manager agent |
rpc.mooons.xyz | RPC gateway, one path per chain | SDKs, generated apps |
events.mooons.xyz | SSE streams and WebSocket rooms | Builder, apps, companion |
hooks.mooons.xyz | Inbound callbacks from Telegram, payment, and identity providers | Outside services |
[project].mooons.site | Production app. Test version at [project]-test | End users |
Colors name cells, not tenants. A cell is a self-contained copy of the API with its own databases. Tenants never choose a color. They call api.mooons.xyz, and the router reads the cell from the API key, such as mk_live_indigo_8f2c…. A bad release or runaway tenant affects one color, and new releases land on a canary cell first.
The API
One address, versioned in the path. This is the proposed shape of the first endpoints, for review, and not yet a specification.
Projects /v1/projects
| GET | /v1/projects | List a tenant's projects | sk |
| POST | /v1/projects | Create a project | sk |
| POST | /v1/projects/:id/releases | Start a release pipeline | sk |
| GET | /v1/projects/:id/events | Read the project event stream | sk |
Wallets /v1/wallets
| POST | /v1/wallets | Create a wallet for a user or agent | pk |
| GET | /v1/wallets/:id | Read addresses and balances | pk |
| POST | /v1/wallets/:id/transactions | Simulate and send through the relay | sk |
| PATCH | /v1/wallets/:id/limits | Set agent spend limits | sk |
Contracts /v1/contracts
| GET | /v1/templates | List audited templates | public |
| POST | /v1/contracts | Deploy a configured template | sk |
| GET | /v1/contracts/:id/audit | Read the audit report | pk |
Example
// react sdk · sketch, not final
import { MooonsProvider, ConnectButton } from "@mooons/react"
<MooonsProvider projectKey="mk_test_indigo_…">
<ConnectButton methods={["email", "passkey", "wallet"]} sponsorGas />
</MooonsProvider>
The stack
Cloudflare is the base for compute, data, and delivery. Everything it does not offer is named as an outside provider or open-source tool. The full list is in the tech stack.
Compute and delivery
| Part | Choice | Kind | Used for |
|---|---|---|---|
| API runtime | Cloudflare Workers, TypeScript | CF | Public API, router, every block |
| API framework | Hono | OSS | Routing, middleware, typed clients |
| Tenant apps | Workers for Platforms | CF | Each app as its own isolated script |
| Custom domains | Cloudflare for SaaS | CF | Tenant domains with automatic certificates |
| Long jobs | Workflows | CF | Pipeline steps that retry, wait, and resume |
| Heavy builds | Containers and Sandbox | CF | Compiling contracts, running tests, building apps |
| Browser automation | Browser Rendering, Playwright | CF | Testing agents using the test version |
Data and state
| Store | Choice | Kind | Holds |
|---|---|---|---|
| Relational | D1 with Drizzle | CF | Tenants, projects, billing, per-project app data |
| Coordinated state | Durable Objects | CF | Transaction queues, SSE, game rooms, agent sessions |
| Cache and config | Workers KV | CF | Cell map, flags, RPC cache |
| Files | R2 | CF | NFT layers, art, artifacts, audit reports, archives |
| Queues | Cloudflare Queues | CF | Webhooks, fan-out, indexer intake, email |
| Vectors | Vectorize | CF | Search over docs, contracts, code, findings |
| Metrics | Workers Analytics Engine | CF | Metering and tracking counts |
| Permanent storage | IPFS pinning, Arweave as an option | Provider | NFT metadata and media |
AI models
Agents are assigned a tier, not a model, so a gateway can swap models without touching any agent. This is a starting map. Benchmark it on real tasks before fixing it.
| Tier | Starting model | API id | Given to |
|---|---|---|---|
| Strongest | Claude Opus 5.5 | claude-opus-5-5 | First builds, contracts, audits, revenue modelling |
| Standard | Claude Sonnet 5.5 | claude-sonnet-5-5 | Edits, deployment plans, insights, manager planning |
| Small | Claude Haiku 5.5 | claude-haiku-5-5 | Chat replies, test steps, classification, summaries |
| Second opinion | A top model from a different provider | Set in the gateway | Verification and audit cross-checks |
Agent runtime
| Part | Choice | Kind | Used for |
|---|---|---|---|
| Model gateway | Cloudflare AI Gateway | CF | Tier routing, fallback, caching, token metering |
| Agent framework | Cloudflare Agents SDK | CF | One long-lived instance per project or user |
| Pipeline engine | Workflows | CF | Build, test, verify, wait, deploy |
| Code sandboxes | Sandbox and Containers | CF | One sandbox per run, discarded after |
| Tool protocol | MCP servers on Workers | Own | Platform tools and one server per app |
| Agent memory | D1, Vectorize, R2 | CF | Project facts, search, transcripts |
| Evaluation | Fixed task set scored on each version | Own | Deciding what leaves the canary cell |
Clients, security, and operations
| Area | Choice |
|---|---|
| Web apps | React, Vite, TanStack Router and Query, Tailwind |
| Desktop and mobile | Tauri 2 with Rive, React Native with Expo |
| SDK packaging | TypeScript, tsup, npm |
| Edge and console | WAF, rate limiting, Turnstile, Cloudflare Access |
| Agent permissions | Scoped keys per run, spend limits on agent wallets |
| Repository and deploys | pnpm, Turborepo, Wrangler from GitHub Actions, canary cell first |
| Tests and errors | Vitest, Playwright, Sentry |
| Status | A hosted page kept off Cloudflare, so it survives an outage there |
Chains
Each chain family needs its own contract templates and its own route to account abstraction. The user experience is the same on all of them: no seed phrase, no gas prompt. Mooons owns one SDK per chain, and each one is tested against that chain's established library. Rollout order is a proposal.
| Family | Mooons SDK | Checked against | Contracts | Account abstraction | Randomness and oracles | Order |
|---|---|---|---|---|---|---|
| EVM | chain-evm | viem | Solidity, Foundry, OpenZeppelin | ERC-4337, EIP-7702, ERC-7579 | Chainlink, Pyth | 1 |
| Solana | chain-solana | @solana/kit | Rust, Anchor | Smart accounts, fee payer | Switchboard, Pyth | 2 |
| TON | chain-ton | @ton/ton, TON Connect | Tact, Blueprint | Wallet v5 gasless transfers | Pyth, commit and reveal | 2 |
| Sui | chain-sui | @mysten/sui | Move | zkLogin, sponsored transactions | Native, Pyth | 3 |
| Aptos | chain-aptos | @aptos-labs/ts-sdk | Move | Keyless accounts, fee payer | Native, Pyth | 3 |
| NEAR | chain-near | near-api-js | Rust, near-sdk | Named accounts, meta-transactions | Native, Pyth | 3 |
Wallets and contracts
One wallet service for people and AI agents. Contracts come from an audited library, so generated apps stay consistent and checkable.
Non-custodial
Keys held by the user. Device enclave and passkeys through WebAuthn.
Semi-custodial
Sign in by email, through a key service where no single party holds the key.
Account abstraction
Smart accounts with sponsored gas, session keys, and spend limits. Also used for agent wallets.
The contract path
If a published template turns out to be vulnerable, a superadmin pulls it, sees every project that deployed it, notifies owners, and approves a migration plan.
Standards to support
Listed from existing notes. Standards change status often, so confirm each against its current specification before committing.
| Standard | What it does | Why Mooons needs it |
|---|---|---|
| EIP-7702 | Lets an ordinary wallet act as a smart account | Existing wallets get batching and sponsored gas |
| ERC-7579 | Modular smart accounts | Session keys, limits, and recovery ship as modules |
| ERC-7715 | Scoped, time-limited wallet permissions | Basis for agent wallets and game session keys |
| ERC-5792 | Batched calls and wallet capabilities | One approval for multi-step actions |
| ERC-7683 | Cross-chain intents | Pay from any chain |
| ERC-6551 | An NFT that owns its own account | Game characters with inventories |
| x402 | Pay-per-request over HTTP in stablecoins | Agents pay for APIs, and tenants can charge for MCP |
| ERC-8004 | Identity and reputation for AI agents, still a draft | Makes agents identifiable to each other |
SDKs
Mooons owns its chain layer, following the structure of OKX's wallet SDK: shared cryptography, one common interface, and one package per chain. Agents and developers learn one interface, and a new chain is a new package rather than a rewrite. Agents never write wallet or chain code from scratch. They compose these packages, which keeps generated apps consistent and auditable.
@mooons/crypto audited primitives: curves, hashes, encodings └─ chain-base one interface every chain implements, test vectors ├─ chain-evm phase 1 ├─ chain-solana phase 3 ├─ chain-ton phase 3 ├─ chain-sui phase 4 ├─ chain-aptos phase 4 └─ chain-near phase 4 wallet-sdk wallet types, signers, agent permissions. Uses any chain package └─ js-sdk apis, auth, util helpers, vanilla js ├─ react-sdk providers, hooks, tanstack query clients │ ├─ gram-sdk telegram mini apps │ └─ base-sdk coinbase mini apps ├─ game-sdk rooms, inventory, quests, markets (recommended addition) └─ mcp-sdk platform tools for coding agents
Proposed dependency order. Confirm whether gram and base should also work without react.
The shared interface
// sketch, not final
import { EvmChain } from "@mooons/chain-evm"
const chain = new EvmChain({ network: "base" })
const address = await chain.getAddress(signer)
const tx = await chain.buildTransaction({ to, value, data })
const signed = await chain.signTransaction(tx, signer)
// Every chain package has these same methods
// Sending goes through the relay and rpc.mooons.xyz
What each layer does
| Layer | Does | Does not do |
|---|---|---|
@mooons/crypto | Wraps audited curve, hash, and encoding libraries behind one API | Implement cryptographic primitives |
chain-base | Defines addresses, transactions, messages, signers, and errors | Contain anything chain-specific |
chain-* | Derives addresses, builds transactions, and signs transactions and messages for one chain | Call the network or hold keys |
wallet-sdk | Adds wallet types, smart accounts, and agent permissions | Know how one chain encodes a transaction |
What we keep from OKX, and what we change
Keep
- Three layers: crypto, a common base, and one package per chain.
- Offline signing: chain packages work without a network.
- Install only what you use: an EVM app never loads TON code.
- Public scope: published on npm as
@mooons/*.
Change
- Signers, not private keys: OKX's example passes a key in. Mooons accepts a signer, so email wallets never rebuild a whole key.
- Account abstraction first: smart accounts and sponsored gas are part of the design.
- Built to run anywhere: Workers, browsers, and React Native, with no Node dependencies.
- Readable by agents: strict types and plain errors, so coding agents use them correctly.
Why own them
| Gain | Cost | How we limit the cost |
|---|---|---|
| One interface across six chains | Six packages to maintain | Ship by phase, and add a chain only when a phase needs it |
| Control over bundle size and runtimes | More work in Phase 1 | Start with EVM and only the transaction types it needs |
| Agents build on a stable surface | Chain upgrades become our job | The Trends agent tracks chain changes, and tests run on each release |
| Fewer outside dependencies in signing | Bugs here can lose funds | Audited primitives, published test vectors, and an outside audit before mainnet |
Testing rule. The established library for each chain stays in the test suite. Every package must produce byte-identical addresses and signed transactions to that library across its test vectors. A difference blocks the release.
Scope. Own SDKs cover the client side. Contract tooling such as Foundry and Anchor, and bundlers, RPC providers, and oracles stay as they are in the Stack tab.
The accelerator
The third ring. A cohort program that takes teams from first build to real users on Mooons, measured on what users actually do. Everything in this tab is a proposal: none of it appears in the earlier plans, and the numbers are starting points.
Why it belongs here
- Mooons can see the numbers: the tracking service already measures usage, so progress is not self-reported.
- The cost is low: most of the support is agent work the platform already runs.
- It proves the product: each launch is a public reference app.
- It feeds the go-to-market: agencies and teams bring their clients and users.
What it is not
- A fund or investment product
- A token launchpad
- A promise of funding or returns
- A route around the release pipeline or audit gate
- A substitute for formal audits on high-value contracts
How a cohort runs
Twelve weeks and about ten teams, reviewed against the same numbers each week.
What each side does
- Apply with a test version Team
The Free plan covers it, so the application doubles as proof that the team can use the platform.
- Selection Admin
People review every application. Agents can summarize one, but do not score or rank teams.
- Allowances switched on Agent
Scale-level allowances and a capped gas grant are applied to the team's tenant.
- Human contract review Admin
A person reviews contracts on top of the AI audit before the first mainnet release.
- Weekly report Agent
Insights and manager agents report active users, retention, transactions, and revenue, with suggested changes.
- Team decides Team
The two human release steps stay with the team. The program never approves or deploys for them.
- Demo day and handover Team
Teams show live numbers. Allowances end, and each team moves to a normal plan with its app untouched.
What teams get
Allowances
Scale plan limits for the cohort, including 2,000 agent credits a month and 25 production apps.
Gas grant
A capped balance for deployment and sponsored transactions. Daily caps still apply.
Human review
People review contracts and the riskiest parts of each release.
Introductions
Users, partners, and investors. Introductions only, with no promised outcome.
Selection criteria
| Criterion | What we look for |
|---|---|
| A working test version | Built on Mooons before applying |
| A reachable audience | A clear way to reach the first users |
| Fit with the stack | Launch samples or wave A. Other types by exception |
| Commitment | A team able to ship and iterate through the 12 weeks |
| Lawful and low risk | No betting, securities, or wave C until those decisions are made |
Measures and cost
Progress comes from the tracking service from Cohort 1 onward, and the program is judged on the same data.
| Measure | Why it matters |
|---|---|
| Time to mainnet | Shows whether the platform does what it promises |
| Monthly active users | Real use, counted the same way as billing |
| Weekly retention | Separates a launch spike from a product |
| Revenue or transaction volume | Whether the app can sustain itself |
| Teams on a paid plan after the cohort | The main test of the program's value |
| Critical findings after launch | Target: none |
Cost of one cohort
Built from the plan's full-use cost of a Scale tenant. Those figures are assumptions until measured in beta. Most teams will not use every allowance.
| Line | Per team, 3 months | 10 teams |
|---|---|---|
| Plan value given (not a cost) | $897.00 | $8,970.00 |
| Agent model spend, $80.00 a month | $240.00 | $2,400.00 |
| Cloudflare, $113.88 a month | $341.64 | $3,416.40 |
| RPC and key service, $20.00 a month | $60.00 | $600.00 |
| Gas grant, assumed | $500.00 | $5,000.00 |
| Cost excluding people | $1,141.64 | $11,416.40 |
Not included: staff time and any outside auditors. Human contract review is likely to be the largest cost, so get a quote before setting the cohort size.
Accelerator track
| Phase | Step | Done when |
|---|---|---|
| 1 | Write terms, selection rules, and review process | Legal review complete |
| 2 | Cohort 0: own brands and partner agencies, with numbers reported by hand because tracking arrives in Phase 3 | Teams launch on mainnet with little Mooons staff help |
| 3 | Cohort 1: first public cohort, using tracking and insights | Most teams stay on a paid plan afterward |
| 4 | Repeat cohorts, with game and mini app tracks | Cost per retained team is known and acceptable |
Pricing
$19 stays as the headline price. A single flat plan cannot hold, because agent tokens, user traffic, and gas all grow without limit. Every figure here is a starting proposal.
| Free | Builder | Pro | Scale | Enterprise | |
|---|---|---|---|---|---|
| Price per month | $0 | $19 | $79 | $299 | Custom |
| Agent credits | 30 | 200 | 800 | 2,000 | Custom |
| Extra credits | None | $10 per 100 | $9 per 100 | $8 per 100 | Volume |
| Monthly active users | 100, test | 1,000 | 10,000 | 50,000 | Custom |
| API and RPC requests | 100,000 | 1 million | 10 million | 50 million | Custom |
| Production apps | None | 1 | 5 | 25 | Unlimited |
| Team seats | 1 | 1 | 5 | 20 | Unlimited |
| Agents | Builder, testing | Adds contract, verification, deployer, manager | Adds insights, gamification | Adds priority runs, custom review | Dedicated capacity |
| API cell | Shared | Shared | Shared | Dedicated as add-on | Dedicated |
The three meters
Agent credits
One credit is a fixed amount of model spend, sold at about 2.5 times its cost. Out of credits, agents pause and live apps keep running.
Monthly active users
A user who signs in or signs a transaction, counted once across production apps. Test users never count.
Gas
Passed through at cost plus 7%. Testnet gas is free, and deployment is quoted before the toggle.
Does each plan hold up at full use?
| Per tenant per month | Builder | Pro | Scale |
|---|---|---|---|
| Revenue | $19.00 | $79.00 | $299.00 |
| Agent model spend | $8.00 | $32.00 | $80.00 |
| Cloudflare | $3.06 | $22.43 | $113.88 |
| RPC, keys, card fees | $2.35 | $7.59 | $28.97 |
| Left over | $5.59, 29% | $16.98, 21% | $76.15, 25% |
Cloudflare at published rates checked 12 October 2026. Model, RPC, key, and card costs are assumptions to measure in beta. Agent spend is the largest line.
Roadmap
Each phase ships something usable and sets up the next. A phase ends when its exit condition is true, not on a date.
Foundation
- Authentication, email, and public API
- Wallet service with email sign-in and EVM account abstraction
- Transaction relay and RPC gateway
- crypto, chain-base, and chain-evm packages
- wallet, js, and react SDKs
- EVM contract deployment with first audited templates
- Subscription billing
- Accelerator terms and selection rules
The builder
- Coding agents with chat and live preview
- Web3 app and generative NFT samples
- Release pipeline, test and production, approval and toggle
- Superadmin console, webhooks, SSE
- AI audit pipeline, $19 plan goes live
- Accelerator Cohort 0: own brands and partner agencies
Agents and reach
- mcp SDK and a generated MCP server per app
- Manager agent, trends feed, indexer
- Tracking service with insights agents
- Telegram bots, chain-solana, chain-ton, gram and base SDKs
- Mobile wallet and first desktop companion
- Accelerator Cohort 1: first public cohort
Games and full multichain
- Game engine: quests, economy, multiplayer, then markets
- game SDK, samples, and gamification agents
- chain-sui, chain-aptos, and chain-near
- Agent wallets for third-party AI agents
- Companion voice and mobile app discovery
- Repeat cohorts with game and mini app tracks
Go to market
Start where builders and apps already exist.
Our own brands as proof
Rebuild existing products on Mooons as public reference apps.
Agencies first
Web3 studios ship client work faster and bring their clients with them.
Telegram mini apps
The clearest gap: non-developers who want a mini app with a wallet.
Samples and accelerator
Every sample is a demo and a starting point, and every cohort adds public launches.
Risks
What could go wrong, and what limits the damage.
| Risk | Impact | Mitigation |
|---|---|---|
| Generated contract loses funds | Severe | Audited library, separate verifier, owner approval, human review for cohorts |
| Bug in our own signing code | Severe | Audited primitives only, test vectors, comparison tests, outside audit before mainnet |
| Key service compromise or failure | Severe | No single party holds a key, key export, compare providers carefully |
| Phishing or scam apps | High | Guard agent, separate tenant domain, takedown workflow |
| Agent spend exceeds revenue | High | Hard caps, tiered models, measure before setting prices |
| Regulation of tokens, betting, or custody | High | Waves, region blocking, review, and legal advice before launch |
| Dependence on one vendor | Medium | Data exports, status page off Cloudflare, cells |
| Accelerator distracts from the core build | Medium | No public cohort until the Phase 2 exit condition is met |
Open decisions
Each needs an owner and an answer before the work it blocks.
Credits only, equity, or token allocation. Recommendation: credits only for Cohorts 0 and 1.
In-house reviewers or a partner audit firm. This sets the largest accelerator cost.
Where key shares live and what recovery looks like. This also determines regulatory exposure.
An AI review is not a formal audit. Wording, report format, and liability must be clear.
Whether agents may write new contract logic, and what review it needs before mainnet.
Whether Mooons hosts betting, prediction markets, token sales, paid loot boxes, and real-world assets.
Whether Mooons replaces or builds on Hashkit, integrates or rebuilds Gumly, and rebuilds owned brands.
How mobile and email wallets relate, and whether the mobile wallet serves builders or all end users.
Which contract actions an agent can take alone, and which need a signed owner approval.
Whether users can export apps, and who pays for mainnet deployment.
Whether the chain packages are open source. Recommendation: MIT, since signing code earns trust by being readable.
Key service, model mapping, and indexer, then credit price and allowances from beta data.
Legal
Issues to take to counsel, not legal advice. These decide what Mooons can offer and where.
Custody
How email wallet keys are held determines whether Mooons counts as a custodian in some places. Decide before Phase 1.
"AI pre-audited"
Avoid implying a formal audit. Set the wording, report format, and liability terms.
Betting and prediction markets
Real-money betting needs a gambling licence in most countries, and prediction markets count as derivatives in some.
Token launches
A token sold to fund an app can be a security. Decide what agents may recommend.
Loot boxes
Randomized rewards can fall under gambling or securities rules. Set allowed reward types and regions.
Real-world assets
Fractional property and most yield-bearing tokens are securities in most countries.
Accelerator terms
Equity or tokens could bring securities and investment rules. Credits-only cohorts avoid most of this. Teams keep their code and apps.
Accelerator conflicts
Selecting and promoting teams can look like an endorsement. State that neither selection nor review guarantees safety or returns.
Glossary
Terms used across this plan.
- Tenant
- A paying account: one person or a team.
- Project
- One app with its contracts, wallets, domain, bot, and MCP server.
- Owner
- The person who approves releases and switches production on.
- Cell
- An isolated copy of the API and its data, named by a color.
- Test version
- A working release on testnet at the project's test address.
- Release record
- The one shared record each agent reads from and writes to.
- Agent credit
- A fixed amount of model spend, used to meter agent work.
- Active user
- An end user who signs in or signs a transaction in the month.
- AI pre-audited
- Reviewed by an automated audit pipeline. Not a formal audit.
- MCP
- Model Context Protocol, which lets agents use an app's tools.
- Signer
- Anything that can sign for a wallet: a device key, a passkey, the key service, or an agent wallet.
- Cohort
- A group of teams going through the accelerator together.
Source documents
The earlier plans hold detail this page leaves out.
| File | Covers |
|---|---|
| landing.html | Public landing page |
| mooons-plan.html | Plan v0.12: domains, services, pricing workings, superadmin workflows |
| mooons-stack.html | Stack v0.1: every technology by layer |
| mooons-agents.html | Agents v0.1: handoffs, permissions, quality measures |
| mooons-app.html | Builder app prototype |
| mooons-deck.html | Pitch deck. Not yet updated for the accelerator |
../PLAN.md | Short written plan |
Brand assets
Three linked rings: the "ooo" in mooons, and the three pillars. The mark is monochrome for small sizes. The gradient belongs to the logo alone, so it is the only colour on a Mooons page.
Isometric mark
The pixel mark built from cubes, on the 30° isometric plane used by the harbor and the voxel moon. Monotone only, for large placements such as covers and section art.
Pixel mark
The same three rings on a pixel grid, in the style of the agent mascots. Monochrome only. Keep it at whole multiples of 47 by 17 so the pixels stay even.
Where each logo goes
| File | Use |
|---|---|
logos/mooons-mark.svg | Navigation and small sizes |
logos/mooons-icon-dark-black.svg | Favicon and app tiles |
logos/mooons-icon-dark-nobox.svg | Hero and large brand moments, on black or white |
logos/mooons-iso-dark.svg | Isometric mark on dark backgrounds |
logos/mooons-iso-light.svg | Isometric mark on light backgrounds |
logos/mooons-pixel-dark.svg | Pixel mark on dark backgrounds |
logos/mooons-pixel-light.svg | Pixel mark on light backgrounds |
Colour
#2f6bff#3a2bff#8b4dff#d98bff#f6b0ff#000000Typography
Headings, weight 500
Body and interface
Code and technical labels
Usage
Do
- Keep pages black, white, and grey
- Write "mooons" in the logo and "Mooons" in sentences
- Use the mark below 32 pixels
Don't
- Use the gradient on buttons, text, or backgrounds
- Stretch or redraw the rings
- Place the gradient logo on a colored background