hermes-alpha

Can a stock AI agent — given nothing but a mission brief — bootstrap an autonomous bug bounty system from scratch?

┌──────────────────────────────────────────────────────────────────┐ │ │ │ Creator (You) │ │ └─ browser terminal / Telegram │ │ └─ Overseer (persistent, strategic — builds the system)│ │ └─ Hunter (ephemeral, tactical — finds the bugs) │ │ └─ subagents (parallel analysis workers) │ │ │ └──────────────────────────────────────────────────────────────────┘

Most AI agent projects give the agent a mountain of custom tools, structured APIs, and carefully engineered infrastructure. Hermes Alpha asks: what if you gave it nothing?

• Builds the Hunter's codebase — security skills, system prompt, tools, Dockerfile
• Deploys the Hunter to its own Fly.io machine
• Monitors the Hunter's performance via logs and Elephantasm memory streams
• Intervenes when it spots problems — soft (runtime guidance injection) or hard (code changes + redeploy)
• Learns which interventions work over time, compounding improvements
• Authenticate with your password
• A WebSocket spawns hermes chat inside a PTY
• Bidirectional I/O streams between your browser and the agent's terminal