SolAegis
Autonomous Agentic Wallet Infrastructure for AI Agents on Solana.
Overview
The Problem
AI agents today are largely read-only, interacting with APIs and social media. To be true economic participants, they need more than just data—they need secure, programmatic control over their own assets. Traditional wallets require human signatures, breaking the autonomous loop.
The Solution
SolAegis is a production-grade runtime that provisions independent, AES-encrypted wallets for AI agents, allowing them to parse natural language, schedule tasks, and execute complex DeFi actions with zero human intervention, tightly controlled by a 10-layer Policy Sandbox.
Architecture
┌──────────────────────────────────────────────────────────────┐
│ FRONTEND / CLI │
│ Next.js Dashboard + React Chat UI + PowerShell CLI │
└───────────────┬──────────────────────────────┬───────────────┘
│ │
▼ ▼
┌──────────────────────────────────────────────────────────────┐
│ AGENT EXECUTION CORE │
│ │
│ ┌─────────────────┐ ┌─────────────────┐ ┌──────────┐ │
│ │ Intent Parser │ → │ Policy Sandbox │ → │ Wallet │ │
│ │ (LLM / Strict) │ │ (10 Layers) │ │ Manager │ │
│ └─────────────────┘ └─────────────────┘ └──────────┘ │
│ │ │
│ ┌─────────────────┐ ┌─────────────────┐ │ │
│ │ SKILLS.md │ │ BullMQ Scheduler│ │ │
│ │ (Capabilities) │ │ Redis Backed │ │ │
│ └─────────────────┘ └─────────────────┘ │ │
└─────────────────────────────────────────────────────┼────────┘
│
┌─────────────────────────▼────────┐
│ SOLANA / ORCA WHIRLPOOL │
└──────────────────────────────────┘Intent Engine
Strict dual-parser system using regex for deterministic matches, falling back to LLM for complex multi-intent instructions.
Policy Sandbox
10 layers of security including role-based access control, spend limits, prompt injection protection, and transaction simulation.
Wallet Manager
Provisions secure AES-256 encrypted custodial wallets per agent, signing transactions definitively without human intervention.
How It Works
1. Multi-Agent Dashboard
Monitor multiple autonomous agents simultaneously. Each agent has its own executing context, risk profile, and AES-encrypted Solana wallet.
2. Natural Language Control & Action Execution
Agents interpret flexible instructions like "swap 0.1 SOL to USDC" or "send 0.5 SOL to TraderBot", automatically resolving agent names to addresses.
3. Simultaneous Multi-Agent CLI
A robust developer CLI allows for running multiple agents concurrently, proving independent execution, policy enforcement, and scalability.
Key Bounty Requirements Reached
Programmatic Key Provisioning
Agents dynamically generate and manage their own isolated Solana wallets. Keys are stored encrypted using AES-256 and never leave the backend environment.
Zero-Intervention Signing & Policy Sandbox
A deterministic execution engine signs transactions autonomously, but only after passing through a 10-layer Policy Sandbox. This protects against prompt injections, prevents overspending, and blocks unapproved actions (e.g., a "monitor" agent cannot execute swaps).
Dynamic SKILLS.md Architecture
Agents do not rely on hardcoded scripts. They hot-reload their capabilities in real-time by reading the SKILLS.md operating manual, instantly learning new DeFi protocols or rules without codebase changes.