Security-First Architecture
RINNAI is built on a non-custodial foundation with multi-layer security and deterministic enforcement. Your keys, your funds, your control.
Security Layers
Multi-layer protection from intent to execution
Non-Custodial Architecture
RINNAI never holds your private keys or controls your funds. All transactions require your signature. You maintain complete custody at all times.
Policy Enforcement
Policies are enforced immutably on-chain via Solana programs. Constraints are validated before execution and cannot be bypassed by off-chain components.
PDA-Based Security
Program Derived Addresses ensure only authorized users can modify policies. PDAs provide cryptographic proof of ownership and authorization.
Complete Auditability
Every intent, plan, policy check, and execution is recorded on-chain. Full transparency enables verification by anyone at any time.
Threat Model
How RINNAI mitigates potential attack vectors
Compromised AI Agent
HighOn-chain policy enforcement prevents execution outside bounds. Even if AI is compromised, it cannot exceed user-defined limits.
Malicious Intent Parsing
MediumUser reviews and signs all transactions. Policy validation occurs both off-chain and on-chain before execution.
Replay Attacks
MediumTransaction nonces and timestamps prevent replay. Each execution is unique and verifiable on-chain.
Policy Manipulation
HighPolicies are hashed and stored in PDAs controlled by user. Only user can modify policies through signed transactions.
Non-Custodial Architecture
What Does Non-Custodial Mean?
Non-custodial means you always control your private keys and funds. RINNAI agents never have access to your keys. All transactions are constructed off-chain and require your signature before submission.
How It Works
Agents generate execution plans and present them to you for approval. When you sign a transaction, it's submitted to the on-chain program which enforces your policy constraints. At no point can an agent access your funds without your explicit signature.
Deterministic Boundaries
Even after signing, the on-chain program enforces deterministic boundaries. If market conditions change or execution would violate your policy, the transaction fails safely. This prevents front-running, sandwich attacks, and unexpected outcomes.
Security Best Practices
- Always verify contract addresses through official channels
- Review transaction details before signing
- Set conservative spending limits initially
- Use program allowlists to restrict interactions
- Enable notifications for all executions
- Regularly review audit logs
- Keep recovery phrase secure and offline
- Test with small amounts first
Responsible Disclosure
If you discover a security vulnerability, please report it to us immediately. We take all reports seriously and will respond within 24 hours.
security@rinnai.xyz