展示HN:AI代理在Gemini的免费套餐下运营我的单人公司 – 每月$0

2作者: ppcvote28 天前原帖
我是一名在台湾的独立开发者。我为我的科技公司构建了四个AI代理,分别处理内容、销售线索、安全扫描和运营——全部使用Gemini 2.5 Flash的免费套餐(每天1,500次请求)。我每月使用约105次。LLM的月费用为0美元。 架构:四个代理在OpenClaw(开源)上运行,使用WSL2在家中运行,配备25个systemd定时器。 它们每天的工作内容: - 在各个平台上生成8条社交媒体帖子(质量控制:生成 → 自我审核 → 如果得分低于7/10则重写) - 与社区帖子互动并自动回复评论(上下文感知,最多2轮) - 通过RSS + HN API + Jina Reader进行研究 → 将情报反馈到内容中 - 运行UltraProbe(AI安全扫描器)进行潜在客户生成 - 监控7个端点,标记过时的线索,同步客户数据 - 当我进行git push时,自动将博客文章发布到Discord(0 LLM令牌——直接使用提交信息) 令牌优化技巧:代理从不进行长时间对话。每个请求都是(1)读取预计算的情报文件(本地markdown,0令牌),(2)一个集中提示,注入所有上下文,(3)一个响应 → 解析 → 行动 → 完成。研究管道(RSS、HN、网页抓取)消耗0 LLM令牌——这只是纯HTTP + Jina Reader。LLM仅用于创意/分析工作。 真实数据: - 27个自动化的Threads账户,超过12K粉丝,超过3.3M浏览量 - 25个systemd定时器,62个脚本,19个情报文件 - RPD利用率:7%(105/1,500)——剩余93%的余量 - 月费用:0 LLM + 约5美元的基础设施费用(Vercel爱好者版 + Firebase免费) 出现的问题: - 7天内产生了127美元的Gemini账单。创建了一个来自启用计费的GCP项目的API密钥,而不是AI Studio。考虑到令牌($3.50/百万)没有速率上限。教训:始终直接从AI Studio创建密钥。 - 互动循环错误:迭代了所有帖子,而不是前N个。一天内消耗了800 RPD,导致其他一切都无法运作。 - Telegram健康检查调用getUpdates,与网关的长轮询冲突。3分钟内出现18条重复消息。 该网站([https://ultralab.tw](https://ultralab.tw))是完全双语的(zh-TW/en),有21篇博客文章,确实——国际化、博客发布和Discord通知都是自动化流程的一部分。 实时代理仪表板:[https://ultralab.tw/agent](https://ultralab.tw/agent) 技术栈:OpenClaw、Gemini 2.5 Flash(免费)、WSL2/systemd、React/TypeScript/Vite、Vercel、Firebase、Telegram Bot、Resend、Jina Reader。 GitHub(操作手册):[https://github.com/UltraLabTW/free-tier-agent-fleet](https://github.com/UltraLabTW/free-tier-agent-fleet) 欢迎就架构、令牌预算或作为一人公司全天候运行AI代理的实际体验提问。
查看原文
I&#x27;m a solo dev in Taiwan. I built 4 AI agents that handle content, sales leads, security scanning, and ops for my tech agency — all on Gemini 2.5 Flash free tier (1,500 req&#x2F;day). I use ~105. Monthly LLM cost: $0.<p>Architecture: 4 agents on OpenClaw (open source), running on WSL2 at home with 25 systemd timers.<p>What they do every day:<p>- Generate 8 social posts across platforms (quality-gated: generate → self-review → rewrite if score &lt; 7&#x2F;10) - Engage with community posts and auto-reply to comments (context-aware, max 2 rounds) - Research via RSS + HN API + Jina Reader → feed intelligence back into content - Run UltraProbe (AI security scanner) for lead generation - Monitor 7 endpoints, flag stale leads, sync customer data - Auto-post blog articles to Discord when I git push (0 LLM tokens — uses commit message directly)<p>The token optimization trick: agents never have long conversations. Every request is (1) read pre-computed intelligence files (local markdown, 0 tokens), (2) one focused prompt with all context injected, (3) one response → parse → act → done. The research pipeline (RSS, HN, web scraping) costs 0 LLM tokens — it&#x27;s pure HTTP + Jina Reader. The LLM only touches creative&#x2F;analytical work.<p>Real numbers:<p>- 27 automated Threads accounts, 12K+ followers, 3.3M+ views - 25 systemd timers, 62 scripts, 19 intelligence files - RPD utilization: 7% (105&#x2F;1,500) — 93% headroom left - Monthly cost: $0 LLM + ~$5 infra (Vercel hobby + Firebase free)<p>What went wrong:<p>- $127 Gemini bill in 7 days. Created an API key from a billing-enabled GCP project instead of AI Studio. Thinking tokens ($3.50&#x2F;1M) with no rate cap. Lesson: always create keys from AI Studio directly. - Engagement loop bug: iterated ALL posts instead of top N. Burned 800 RPD in one day and starved everything else. - Telegram health check called getUpdates, conflicting with the gateway&#x27;s long-polling. 18 duplicate messages in 3 minutes.<p>The site (<a href="https:&#x2F;&#x2F;ultralab.tw" rel="nofollow">https:&#x2F;&#x2F;ultralab.tw</a>) is fully bilingual (zh-TW&#x2F;en) with 21 blog posts, and yes — the i18n, blog publishing, and Discord notifications are all part of the automated pipeline.<p>Live agent dashboard: <a href="https:&#x2F;&#x2F;ultralab.tw&#x2F;agent" rel="nofollow">https:&#x2F;&#x2F;ultralab.tw&#x2F;agent</a><p>Stack: OpenClaw, Gemini 2.5 Flash (free), WSL2&#x2F;systemd, React&#x2F;TypeScript&#x2F;Vite, Vercel, Firebase, Telegram Bot, Resend, Jina Reader.<p>GitHub (playbook): <a href="https:&#x2F;&#x2F;github.com&#x2F;UltraLabTW&#x2F;free-tier-agent-fleet" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;UltraLabTW&#x2F;free-tier-agent-fleet</a><p>Happy to answer questions about the architecture, token budgeting, or what it&#x27;s actually like running AI agents 24&#x2F;7 as a one-person company.