Provisions secure, non-custodial Turnkey enclaves enabling AI agents to hold SOL, execute transactions, and stake autonomously.

Autonomous AI agents operating on-chain face a fundamental security dilemma: storing private keys in application memory exposes funds to memory leaks and key theft, while requiring manual human approvals destroys agent autonomy.
I built Karen as an open-source agentic wallet runtime using Turnkey Secure Enclaves:
Karen programmatically provisions Solana keypairs inside Turnkey Hardware Security Modules (HSM). The AI agent executes transactions using cryptographic signing credentials without ever exposing raw private keys in memory.
Enforces execution boundaries prior to signing transactions:
Terminal Workspace// Transaction Guardrail Policy Validator export function validateTransactionPolicy(tx: Transaction, policy: PolicyRules): boolean { if (tx.amount > policy.maxSpendPerTx) return false; if (!policy.allowedPrograms.includes(tx.programId.toBase58())) return false; return true; }
Supports automated Devnet validator staking delegation, SPL token wrapping, and account rent reclamation, paired with a web control dashboard.
Programmatically provisioned non-custodial Turnkey enclave wallets for autonomous AI agents.
Automated payload-driven transaction signing without requiring human approval bottlenecks.
Enabled native Solana validator staking, token swaps, and account rent reclamation for agents.
Constructed a real-time web control dashboard for monitoring agent balances and transaction guardrails.