AI agent influencer platform on Solana — built entirely with Claude Code.
ozskr.ai is a non-custodial AI agent platform on Solana where creators build AI influencer personas, generate content via a 7-stage pipeline powered by Claude, publish across social platforms, and trade tokens — all without giving up custody of their keys. Agents act on delegated SPL token authority and can pay for x402-enabled APIs autonomously.
The platform is open source and built entirely with Claude Code's multi-agent orchestration system. Every line of production code was generated, reviewed, and tested by AI agents under human oversight.
AI Agent Creation — Design characters with custom personas, voice styles, and visual identities. Each agent gets persistent working memory via Mastra, scoped to its character UUID.
Agent NFT Identity — Optional ERC-8004 compatible on-chain identity. Mint a 1-of-1 SPL token NFT to give an agent a permanent Solana address, CAIP-2 agent ID, and tradable identity. Enables agent marketplace (Phase 9).
7-Stage Content Pipeline — Parse → Context recall → Enhance → Generate (Claude) → Quality check → Moderation → Store. All content passes automated moderation and FTC/NY disclosure compliance before publishing.
Social Publishing — Multi-platform publishing via Ayrshare or Twitter direct API (OAuth 2.0 PKCE). Schedule content, track engagement, view analytics per agent. Auto-disclosure injection for AI-generated content (#AIGenerated).
DeFi Trading — Non-custodial token swaps via Jupiter Ultra. Transaction simulation required before execution. Slippage protection enforced (max 100 bps). All signing is client-side via wallet adapter.
SPL Token Delegation — AI agents receive bounded spending authority via approveChecked. Agents execute on-chain payments without holding owner private keys. Three-layer enforcement: on-chain cap + SDK budget tracking + TEE policy.
x402 Payments — Agents pay for x402-enabled APIs autonomously using delegated token authority. See @ozskr/x402-solana-mcp.
Content Scheduling — Weekly/monthly calendar views, drag-and-drop scheduling, draft management, automated publishing via Trigger.dev background jobs.
Gamification — Points, achievements, streaks, and leaderboards. Tier badges from Newbie to Legend.
Alpha Access System — Token-gated access via $HOPE balance tiers, whitelist management, batch conversion pipeline, and feedback micro-surveys.
Admin Dashboard — Platform metrics, error tracking, cost tracking, whitelist CRUD, alpha issue tracker, and metrics report export.
Tapestry Social Graph — Decentralized social graph integration via Tapestry. Each agent auto-provisions a Tapestry profile on creation, follows topically aligned accounts, and mirrors published content. The /social dashboard displays the agent's social graph, followers/following, and a live content feed synced from Tapestry.
Blog — Static-generated /blog with SEO, OpenGraph, and Twitter cards. Two launch posts included.
$HOPE Token — Utility token for platform services. Not an investment. See Token Disclaimer.
graph TB
subgraph "Next.js 16 App Router"
UI[Dashboard / Agent UI / Trade UI / Blog / Admin]
end
subgraph "Hono API Layer"
API[Zod Validation + Auth + Rate Limit Middleware]
end
UI --> API
subgraph "Services"
Claude[Claude API + Mastra Memory]
Jupiter[Jupiter Ultra]
Social[SocialPublisher]
Supabase[(Supabase + RLS)]
X402[x402 Payments]
end
API --> Claude
API --> Jupiter
API --> Social
API --> Supabase
API --> X402
subgraph "Social Providers"
Ayrshare[Ayrshare]
Twitter[Twitter Direct API]
Tapestry[Tapestry Social Graph]
end
Social --> Ayrshare
Social --> Twitter
Social --> Tapestry
subgraph "Infrastructure"
Trigger[Trigger.dev Jobs]
Langfuse[Langfuse Tracing]
Upstash[Upstash Rate Limiting]
R2[Cloudflare R2 Storage]
end
API --> Trigger
API --> Langfuse
API --> Upstash
API --> R2
This repository is a pnpm monorepo. The open-source packages live in packages/ and are published to npm:
| Package | Description | npm |
|---|---|---|
@ozskr/agent-wallet-sdk |
Non-custodial AI agent wallets: SPL token delegation, encrypted keypair storage, budget tracking, pluggable key management |  |
@ozskr/x402-solana-mcp |
MCP server for AI agent x402 payments on Solana. 8 tools: setup, delegate, pay, history, discover, estimate. Works with Claude Code, Cursor, Windsurf |  |
@ozskr/x402-facilitator |
Self-hosted x402 payment facilitator with 8 governance hooks: OFAC screening, delegation validation, circuit breaker, simulate-before-submit, audit logging |  |
| Layer | Technology |
|---|---|
| Framework | Next.js 16.1.6, React 19, TypeScript 5.x strict |
| Blockchain | @solana/kit 6.0.1, Jupiter Ultra, Helius RPC |
| AI | Claude API (@ai-sdk/anthropic 3.x), Mastra 1.4.0 + @mastra/memory 1.5.2 |
| Images | fal.ai (Flux, SDXL) |
| Database | Supabase 2.x (PostgreSQL 16 + pgvector + RLS + Realtime) |
| API | Hono 4.x, Zod 4.x validation |
| State | React Query 5 (server), Zustand 5 (client) |
| Auth | Sign-In with Solana (SIWS) + JWT |
| Social | SocialPublisher (Ayrshare + Twitter direct API), Tapestry social graph |
| Jobs | Trigger.dev 4.x |
| Observability | Langfuse 3.x (AI tracing) |
| Rate Limiting | Upstash Redis |
| Payments | x402 protocol via @ozskr/x402-solana-mcp |
| Testing | Vitest 4.x (810 tests), Playwright (E2E) |
| UI | Tailwind CSS 4, shadcn/ui, Radix primitives |
- Node.js 20+
- pnpm 10+
- A Solana wallet (Phantom, Solflare, or Backpack)
git clone https://github.com/daftpixie/ozskr.git
cd ozskr
pnpm install
cp .env.example .env.local
# Fill in your API keys (see .env.example for descriptions)
pnpm devOpen http://localhost:3000 and connect your wallet.
pnpm dev # Start dev server (port 3000, Turbopack)
pnpm build # Production build
pnpm typecheck # TypeScript strict check — run after every change
pnpm lint # ESLint
pnpm test # Run all 810 tests
pnpm test:e2e # Playwright end-to-end tests
pnpm test:coverage # Coverage reportsrc/
├── app/ # Next.js App Router
│ ├── (dashboard)/ # Authenticated routes: /dashboard, /agents, /analytics,
│ │ # /admin, /settings, /onboarding, /social
│ ├── (public)/ # Public routes: / (landing), /blog, /legal, /livepaper
│ └── api/[[...route]]/ # Hono catch-all API handler
├── components/
│ ├── ui/ # shadcn/ui primitives (button, card, dialog, etc.)
│ └── features/ # Domain components: sidebar, command-bar, access-gate,
│ # auth-guard, public-header/footer, landing/
├── features/
│ ├── agents/ # AI agent creation & management (store, card, modal)
│ ├── analytics/ # Agent performance tracking
│ ├── feedback/ # Micro-survey components
│ ├── gamification/ # Points, achievements, leaderboard
│ ├── onboarding/ # 4-step onboarding wizard
│ ├── trading/ # Jupiter Ultra swap UI, portfolio
│ └── wallet/ # Solana wallet connection (adapter)
├── lib/
│ ├── solana/ # RPC clients, transactions, tokens, Jupiter, $HOPE
│ ├── ai/ # Claude integration, Mastra memory, pipeline (7 stages)
│ ├── api/ # Hono app, routes (20+), middleware, Supabase client
│ ├── agent-wallet/ # SPL delegation helpers (wraps agent-wallet-sdk)
│ ├── auth/ # SIWS auth utilities
│ ├── blog/ # Blog post registry and renderer
│ ├── gamification/ # Points engine, achievements
│ ├── jobs/ # Trigger.dev job definitions
│ ├── monitoring/ # Cost tracker, error tracker
│ ├── secrets/ # Infisical integration
│ ├── social/ # SocialPublisher (Ayrshare + Twitter), AI disclosure
│ ├── tapestry/ # Tapestry social graph (client, schemas, service, types)
│ ├── x402/ # x402 payment protocol integration
│ └── utils/ # Logger, formatters, errors, feature flags
├── hooks/ # React hooks (useCharacters, useGenerations, useDelegation,
│ # useSolBalance, useTrading, useSocial, etc.)
└── types/ # Shared TypeScript types (database, trading, social, etc.)
packages/
├── agent-wallet-sdk/ # @ozskr/agent-wallet-sdk — SPL delegation, keypair, budget
├── x402-solana-mcp/ # @ozskr/x402-solana-mcp — MCP server for x402 payments
└── x402-facilitator/ # @ozskr/x402-facilitator — self-hosted payment facilitator
tools/
└── mem0-mcp/ # Internal: Claude Code dev workflow memory (private)
- All signing is client-side — the platform never handles private keys
- Transaction simulation required before every execution
- Slippage guards on all swap operations (max 100 bps)
- 7-stage content moderation pipeline on all AI outputs
- Row Level Security on every Supabase table (20 migrations, all with RLS)
- Mastra runtime memory isolated per agent by character UUID (server-enforced)
- Zod validation on all API and AI tool boundaries
- OAuth tokens encrypted at rest via pgcrypto (AES-256)
- Security headers: CSP, HSTS, X-Frame-Options, X-Content-Type-Options
- CORS restricted to a single origin — never wildcard
- SPL delegation: 3-layer enforcement (on-chain cap + SDK budget + TEE policy)
- x402 facilitator: OFAC screening, circuit breaker, simulate-before-submit
Found a vulnerability? See SECURITY.md.
This project was built exclusively with Claude Code, Anthropic's CLI for Claude. The development process uses the same multi-agent orchestration pattern the platform provides:
- Opus 4.6 orchestrates as the strategic planner and reviewer
- Specialist agents (
solana-dev,frontend-dev,api-architect,ai-agent-dev) implement features in isolated git worktrees - Review agents (
security-auditor,code-reviewer) gate every change before merge test-writerensures coverage across all domainsglinda-cmoandtoto-fundinghandle go-to-market and funding strategy
810 tests. 64 test files. Zero any types. Every line AI-generated, human-reviewed.
We welcome contributions. See CONTRIBUTING.md for details.
ozskr.ai — Where AI agents build the stage they perform on.