⚡ Karen

Autonomous Wallet Infrastructure for Solana AI Agents Karen is an open-source, OpenClaw-inspired agentic wallet runtime where AI agents autonomously manage Solana wallets, sign transactions, and interact with DeFi protocols. It also serves as infrastructure — any external AI agent can plug in via REST API or MCP.

Why Karen?

Karen bridges the gap between AI agents and blockchain by providing:

Autonomous Operation

LLM-powered agents that observe wallet state, reason about decisions, and execute transactions autonomously using pluggable skills

Secure by Design

AES-256-GCM encrypted keystores with HD derivation, per-agent spending limits, rate limiting, and program allowlists

DeFi Native

Built-in Jupiter swaps, native SOL staking, SPL token launches, wSOL wrapping, and token management

Universal Integration

REST API and MCP server let any external AI agent (Claude, OpenClaw, LangChain) access wallet capabilities

Architecture

Karen uses a three-layer architecture that separates agent intelligence, core wallet operations, and blockchain interactions:

┌─────────────────────────────────────────────────────────┐
│         External Agents (OpenClaw, LangChain)           │
└───────────────────┬─────────────────────────────────────┘
                    │
        ┌───────────┴───────────┐
        │     MCP Server        │──── REST API (Port 3001)
        │  17 Wallet Tools      │
        └───────────┬───────────┘
                    │
        ┌───────────┴───────────────────────────┐
        │       Agent Runtime                   │
        │  Observe → Think → Act → Remember     │
        │  • OpenAI / Claude / Grok / Gemini    │
        │  • 17 Pluggable Skills                │
        │  • Persistent Memory                  │
        └───────────┬───────────────────────────┘
                    │
        ┌───────────┴───────────────────────────┐
        │         Core Engine                   │
        │  • Wallet Manager (HD Derivation)     │
        │  • Transaction Engine                 │
        │  • Security Guardrails                │
        │  • Encrypted Keystores (AES-256-GCM)  │
        └───────────┬───────────────────────────┘
                    │
        ┌───────────┴───────────────────────────┐
        │      Protocol Adapters                │
        │  Jupiter • SPL Token • Staking        │
        │  Token Launcher • Wrapped SOL         │
        └───────────┬───────────────────────────┘
                    │
            Solana Devnet RPC

Key Features

🤖 Autonomous Agents

LLM-powered agents (OpenAI, Claude, Grok, Gemini) that follow a continuous observe-think-act loop:

Observe: Check wallet balances, recent transactions, and on-chain state
Think: LLM reasons about strategy and selects the best action
Act: Execute one of 17 pluggable skills (swap, transfer, stake, etc.)
Remember: Persist decisions in memory for learning over time

Agents operate autonomously with configurable loop intervals (default: 30 seconds).

🔐 Secure Wallets

AES-256-GCM encryption: Private keys encrypted with scrypt-derived keys
HD derivation: BIP-44 compatible hierarchical deterministic wallets
Multi-agent support: Each agent gets its own derived wallet from a master seed
Keystore isolation: Encrypted keystores stored separately from application code

💱 DeFi Integration

Jupiter DEX: Automated token swaps with slippage protection
Native staking: Delegate SOL to validators, manage stake accounts
SPL token lifecycle: Create tokens, mint supply, revoke authorities
Token management: Burn tokens, close accounts, freeze/thaw
wSOL wrapping: Convert between SOL and Wrapped SOL

All operations work on Solana devnet for safe testing.

🛡️ Security Guardrails

Every transaction passes through configurable safety checks:

Guardrail	Default	Purpose
`maxSolPerTransaction`	2 SOL	Limit per-transaction spend
`maxTransactionsPerMinute`	5	Rate limiting
`dailySpendingLimitSol`	10 SOL	Daily spending cap
`allowedPrograms`	System, Token, Jupiter, Stake	Whitelist programs
`blockedAddresses`	`[]`	Blacklist addresses

Blocked transactions are logged with reason and never executed.

🔌 Universal Integration

REST API (Port 3001): HTTP endpoints for wallet creation, balance checks, swaps, transfers, staking, and more. MCP Server: Model Context Protocol integration provides 17 tools that any MCP-compatible agent can use:

karen_create_wallet, karen_balance, karen_swap
karen_transfer, karen_airdrop, karen_stake
karen_launch_token, karen_mint_supply, karen_burn
And 8 more DeFi operations

🧠 Agent Skills (17 Total)

Core Skills:

check_balance - View SOL and SPL token holdings
swap - Trade tokens via Jupiter DEX
transfer - Send SOL to addresses
airdrop - Request devnet SOL
token_info - Look up token metadata
wait - Skip action this cycle

Token Lifecycle:

launch_token - Create new SPL token
mint_supply - Mint additional tokens
revoke_mint_authority - Permanently disable minting

Staking:

stake_sol - Delegate to validator
unstake_sol - Deactivate stake account
withdraw_stake - Withdraw deactivated stake
list_stakes - View all stake accounts

Token Management:

burn_tokens - Destroy tokens
close_token_account - Reclaim rent SOL
wrap_sol - Convert SOL to wSOL
unwrap_sol - Convert wSOL to SOL

💾 Agent Memory

Agents maintain persistent memory of past decisions:

// src/agent/runtime.ts:214
this.memory.addMemory(this.config.id, {
  cycle: this.cycle,
  reasoning,
  action: action
    ? `${action.skill}(${JSON.stringify(action.params)})`
    : null,
  outcome,
  timestamp: new Date().toISOString(),
})

Memory is included in future LLM prompts so agents learn from experience.

📊 Dashboard & CLI

CLI: Full-featured terminal interface for wallet/agent management
Dashboard: Premium Next.js UI for live monitoring (optional)
Telegram Bot: Control agents via Telegram (optional)

Supported LLM Providers

Provider	Models
OpenAI	gpt-4o, gpt-4o-mini, gpt-3.5-turbo
Anthropic	claude-sonnet-4-20250514, claude-3-haiku
xAI	grok-3-latest
Google	gemini-2.0-flash

Configured via environment variables. At least one API key required.

Project Structure

karen/
├── src/
│   ├── core/                 # Wallet, transactions, guardrails
│   │   ├── wallet/           # Wallet manager, HD derivation, keystores
│   │   ├── transaction/      # Transaction engine, guardrails
│   │   └── solana/           # RPC connection
│   ├── agent/                # LLM-powered agent runtime
│   │   ├── llm/              # OpenAI, Anthropic, Grok, Gemini providers
│   │   ├── skills/           # 17 pluggable agent skills
│   │   ├── memory/           # Persistent agent memory
│   │   ├── orchestrator.ts   # Multi-agent management
│   │   └── runtime.ts        # Observe-think-act loop
│   ├── protocols/            # DeFi integration
│   │   ├── jupiter.ts        # DEX swaps
│   │   ├── spl-token.ts      # Token operations
│   │   ├── staking.ts        # Native SOL staking
│   │   ├── token-launcher.ts # Token creation
│   │   └── wrapped-sol.ts    # wSOL wrapping
│   ├── api/                  # REST API server
│   ├── mcp/                  # MCP server
│   └── cli/                  # CLI interface
├── dashboard/                # Next.js monitoring dashboard
├── data/                     # Runtime data (keystores, logs, memory)
├── SKILLS.md                 # Agent skill reference
└── .env.example              # Configuration template

Security Model

Karen implements defense-in-depth security:

Encrypted keystores: AES-256-GCM encryption with scrypt key derivation (N=16384, r=8, p=1)
HD wallets: BIP-44 derivation path m/44'/501'/0'/0' for Solana
Spending limits: Per-transaction, per-minute, and daily caps enforced before signing
Program allowlists: Only whitelisted Solana programs (System, Token, Jupiter, Stake, Metadata)
Audit logging: Every transaction and decision logged with timestamp
Devnet only: Designed for safe testing without real funds

Next Steps

Quickstart

Get from zero to running agent in 5 minutes

Installation

Prerequisites, dependencies, and environment setup

API Reference

REST API and MCP tool documentation

Agent Skills

Complete reference for all 17 agent skills

Community & Support

GitHub: github.com/yourusername/karen
Issues: Report bugs and request features
Discussions: Ask questions and share strategies
License: MIT - free for commercial and personal use

Get Started

Core Concepts

Integration

Agent Skills

Development

Introduction

⚡ Karen

Why Karen?

Autonomous Operation

Secure by Design

DeFi Native

Universal Integration

Architecture

Key Features

🤖 Autonomous Agents

🔐 Secure Wallets

💱 DeFi Integration

🛡️ Security Guardrails

🔌 Universal Integration

🧠 Agent Skills (17 Total)

💾 Agent Memory

📊 Dashboard & CLI

Supported LLM Providers

Project Structure

Security Model

Next Steps

Quickstart

Installation

API Reference

Agent Skills

Community & Support

Build docs developers (and LLMs) love

Get Started

Core Concepts

Integration

Agent Skills

Development

​⚡ Karen

​Why Karen?

Autonomous Operation

Secure by Design

DeFi Native

Universal Integration

​Architecture

​Key Features

​🤖 Autonomous Agents

​🔐 Secure Wallets

​💱 DeFi Integration

​🛡️ Security Guardrails

​🔌 Universal Integration

​🧠 Agent Skills (17 Total)

​💾 Agent Memory

​📊 Dashboard & CLI

​Supported LLM Providers

​Project Structure

​Security Model

​Next Steps

Quickstart

Installation

API Reference

Agent Skills

​Community & Support

Build docs developers (and LLMs) love

⚡ Karen

Why Karen?

Architecture

Key Features

🤖 Autonomous Agents

🔐 Secure Wallets

💱 DeFi Integration

🛡️ Security Guardrails

🔌 Universal Integration

🧠 Agent Skills (17 Total)

💾 Agent Memory

📊 Dashboard & CLI

Supported LLM Providers

Project Structure

Security Model

Next Steps

Community & Support