Show HN: I Put an AI Agent on a $7/Month VPS with IRC as Its Transport Layer

Why It Matters

It demonstrates a low‑cost, self‑hosted alternative to commercial AI chat widgets, giving developers granular control over data privacy and operational expenses.

Key Takeaways

• AI agent runs on $7/month VPS with minimal resources
• Uses IRC as transport, avoiding third‑party API lock‑in
• Two-tier model architecture balances cost and capability
• Public nullclaw bot isolates private data on separate ironclaw box
• Strict sandboxing and daily $2 budget limit abuse potential

Pulse Analysis

Traditional portfolio chatbots often rely on feeding a static résumé into a large language model, resulting in generic, re‑hashed answers that add little value. By contrast, the nullclaw/ironclaw setup treats the AI as a real‑time code analyst. When a visitor asks, for example, how test coverage is handled, the bot clones the relevant repository, parses the CI configuration, and returns concrete metrics. Leveraging IRC as the communication layer eliminates dependence on proprietary SDKs, offers a familiar terminal experience, and ensures the entire stack remains under the developer’s control.

The architecture is deliberately split into two security zones. Nullclaw, a 678 KB Zig binary, runs on the cheap VPS and only interacts with public IRC channels, providing greetings and lightweight queries. More sensitive operations—email access, calendar scheduling, or deep code analysis—are delegated to ironclaw, a separate machine reachable only via Tailscale. Cost efficiency comes from a tiered model strategy: Haiku 4.5 handles the hot path with sub‑second latency and pennies per conversation, while Sonnet 4.6 is invoked for heavier tasks, keeping the overall spend below $2 per day. Hardened SSH, a minimal firewall, Cloudflare edge protection, and strict sandboxing further reduce the attack surface.

Beyond the novelty of an IRC‑backed AI assistant, this project signals a broader shift toward self‑hosted, protocol‑agnostic AI services. Developers can now embed intelligent agents without surrendering data to third‑party platforms, and the use of mature, open protocols like IRC guarantees long‑term stability and interoperability. As token costs rise and privacy concerns intensify, solutions that combine cheap compute, modular model selection, and rigorous security—such as this dual‑agent system—are likely to inspire more personalized, cost‑effective AI deployments across the tech industry.