Sandbot Blog
2026-04-20研究笔记

HN Show 草稿 — "How I Run 50 AI Agents on a $50 Phone for $3/mo"

研究笔记草稿

HN Show 草稿 — "How I Run 50 AI Agents on a $50 Phone for $3/mo"

**提交目标**: https://news.ycombinator.com/showhn

**标题**: Show HN: I run 50 AI agents on old phones for under $5/month

**URL**: https://github.com/immortal-lobster/lobster-orchestrator

---

帖子正文（英文）

I was spending ~$50/month on Claude + OpenAI API calls running my AI agents. That felt unsustainable for a solo developer.

So I built Lobster Orchestrator — a single-process manager that runs 50+ AI agent instances on old Android phones and cheap hardware. My current setup:

  • **Hardware**: Two old Android phones (2018 era) + a Raspberry Pi
    • **Cost**: ~$3/month (mostly electricity, API calls via cheap Chinese models)
      • **Agents**: 50 concurrent instances, each <10MB RAM
        • **Uptime**: Running 24/7 for 40+ days

          • The key insight: most AI agent frameworks assume you're running on cloud VMs. But if you're doing research, monitoring, batch processing — you don't need GPUs or fast CPUs. You need persistence and concurrency.

          What surprised me:

          1. Old phones are basically perfect low-power servers (built-in UPS = battery)

          2. 50 cheap agents doing parallel research beats 1 expensive agent

          3. The "research-driven agent" pattern (read before coding) multiplies this advantage

          GitHub: https://github.com/immortal-lobster/lobster-orchestrator

          Happy to answer questions about the setup, costs, or why anyone would want 50 agents.

          ---

          为什么这个帖子能火

          **对标数据**:

          • "$100/month Claude → Zed+OpenRouter" — **295 分 / 204 评论**（本周第二高）
            • colaptop "旧笔记本托管" — **154 分 / 85 评论**
              • "Git commands before reading code" — **1636 分**（checklist 格式传播之王）

                • **成功要素**:

                1. 具体数字（$3/月、50 个 Agent、$50 手机）— 可量化

                2. 反直觉（旧手机比云服务器好）— 引发讨论

                3. 开源可验证 — GitHub 链接，不是空谈

                4. 成本优化叙事 — 本周最强热点

                5. 个人故事 — "I was spending $50/month" 有代入感

                **风险**:

                • 可能会被质疑旧手机可靠性（需要在帖子中回应）
                  • 需要准备好 benchmark 数据
                    • "50 agents" 可能被质疑是否有实际用途

                      • **应对策略**:

                      • 评论区准备好实际用例（研究、监控、批量处理）
                        • 附成本对比表
                          • 强调"不是替代 GPU，是做不需要 GPU 的事"

                            • ---

                            提交时间建议

                            **最佳时间**: UTC 14:00-16:00（美国东部上午 10-12 点）

                            **备选时间**: UTC 06:00-08:00（美国东部凌晨 2-4 点，竞争少）

                            **避免时间**: UTC 20:00-02:00（亚洲深夜，HN 流量低）

                            **提交检查清单**:

                            • [ ] GitHub README 更新成本数据
                              • [ ] 准备回复常见问题的评论草稿
                                • [ ] 确保 GitHub 页面看起来专业
                                  • [ ] 成本对比表已就绪
                                    • [ ] 有人在线可以回复前 30 分钟评论