Three-Layer Memory and Heartbeat Crons as Autonomous Agent Infrastructure
3-layer memory + heartbeat cron + delegation pattern maps directly to Vault memory tiers, Inngest crons, and gateway heartbeat architecture
Nat Eliason gave an OpenClaw bot named Felix $1,000 and told it to build a business. Three weeks later it had made $14,718 — launching its own website, info product, and X account. Peter Yang walked through the full setup in a 35-minute interview.
The architecture is what’s worth paying attention to. Felix runs on a three-layer memory system that separates context into tiers — presumably something like working memory, episodic recall, and long-term knowledge. That layered approach is what lets the agent maintain coherent goals across sessions instead of starting from scratch every time. On top of that, heartbeat crons keep the agent alive and proactive — it doesn’t just respond to prompts, it wakes up on a schedule, checks state, and decides what to do next. And when it needs heavy lifting, it delegates to Codex for implementation work.
The multi-threaded chat setup is also notable — running five parallel projects from one agent by splitting work across separate conversation threads. Combined with prompt injection defenses for its public X presence, the whole thing is less “chatbot with a wallet” and more “autonomous agent with an operational architecture.” The $100K+ in crypto it somehow accumulated is a wild footnote.
What’s interesting is how much of this maps to primitives that already exist in different forms: durable cron functions, layered memory stores, multi-agent fan-out, heartbeat monitoring. The difference is someone wired them together with the explicit goal of economic autonomy — the agent isn’t assisting a human workflow, it’s running its own.
Key Ideas
- Three-layer memory separates agent context into tiers (working / episodic / long-term) enabling coherent behavior across sessions
- Heartbeat crons make agents proactive rather than reactive — wake up, check state, decide next action
- Delegation to Codex for implementation keeps the orchestrator agent focused on decisions
- Multi-threaded chats enable parallel project execution from a single agent identity
- Prompt injection defense is a real concern when an agent has a public-facing X account
- The pattern is: layered memory + scheduled heartbeat + delegation + parallel execution = autonomous economic agent