A tiny CFO sitting on your taskbar蹲在任务栏上的小小 CFO
Dollar reads your local AI logs and the API calls it makes for you, then turns them into a running picture of what your coding agents cost. One provider is measured exactly; the others are careful estimates. We are honest about which is which.Dollar 读取本地的 AI 日志，以及它替你发出的 API 调用，把这些拼成一张实时的成本图——你的编程助手到底花了多少钱。其中一家是精确计量，其余是审慎估算。哪个是哪个，我们都如实标注。
Exact where we can, estimated where we must能精确就精确，该估算就估算
Not all spend is knowable the same way. Dollar labels every number so you always know whether you are looking at a measurement or a model.不是所有花费都能用同一种方式得知。Dollar 给每个数字都打上标签，你随时都清楚自己看的是实测值，还是推算值。
For OpenAI calls made through Dollar, we count the real tokens returned by the API. This is the one number we measure rather than estimate — your actual usage, model by model.对于经 Dollar 发出的 OpenAI 调用，我们按 API 实际返回的 token 计数。这是唯一一个实测、而非估算的数字——逐个模型，对应你的真实用量。
Dollar reads Claude Code's local session logs and prices the tokens — including cache reads and writes — against an editable per-model rate table. It is a close estimate, not your bill; actual billing may differ.Dollar 读取 Claude Code 的本地会话日志，按一张可编辑的分模型价目表为 token 计价——包括缓存读取与写入。这是接近的估算，不是你的账单；实际计费可能有出入。
Codex usage is estimated from local logs too. Its price is an editable placeholder you set yourself — so treat Codex figures as a rough guide until you dial in a rate you trust.Codex 的用量同样据本地日志估算。它的单价是一个由你自己填写的可编辑占位值——在你设定一个可信的费率之前，Codex 的数字只能当作粗略参考。
Every rate lives in settings and you can change it. Out of the box Dollar ships prices for 每一条费率都在设置里，你随时可改。开箱即带 gpt-4o · gpt-4o-mini · claude-sonnet · claude-haiku · claude-opus — with separate input, output, and cache-token rates so the math reflects how these models actually charge. 的价格——输入、输出、缓存 token 分别计价，让账算得更贴近这些模型真实的收费方式。
Know your window before you hit the wall在撞上限额之前，先看清窗口
Rate limits reset on a clock. Dollar tracks the windows and counts down to the next reset, so a stall mid-task is never a mystery.限额是按时间窗口重置的。Dollar 会跟踪这些窗口，并倒计时到下一次重置——任务中途卡住，不再莫名其妙。
Dollar follows Claude's 5-hour window and Codex's usage windows, showing a live countdown to the next reset. When you are close to the edge, you see it coming instead of guessing why the agent went quiet.Dollar 跟踪 Claude 的 5 小时窗口和 Codex 的用量窗口，实时倒计时到下一次重置。临近上限时，你能提前看到，而不是猜测助手为什么突然安静了。
By comparing your usage against Pro, Max 5x and Max 20x, Dollar nudges you toward the plan that fits: a possible downgrade when you are using under 35%, a possible upgrade when you push past 85%. Suggestions only — the choice stays yours.通过把你的用量与 Pro、Max 5x、Max 20x 对比，Dollar 会提示更合适的套餐：用量低于 35% 时提醒可考虑降档，超过 85% 时提醒可考虑升档。只是建议——选择权始终在你手里。
A guide rail, not a surprise at month-end一道护栏，而非月底的惊吓
Set a monthly budget and Dollar watches your estimated spend against it — flagging trouble early and pointing at where the money actually goes.设定一个月度预算，Dollar 会盯着你的估算花费——提前预警，并指出钱到底花在了哪里。
Pick a monthly number and Dollar warns you at 80% and again when estimated spend crosses it. It is a local guide against an estimate — not a bill from any provider.填一个月度数字，Dollar 会在到达 80% 时、以及估算花费越线时分别提醒你。它是对照估算值的本地护栏——不是任何供应商寄来的账单。
If today's spend runs around twice your recent daily average, Dollar flags it. A runaway loop or a too-expensive model gets caught the same day, not three weeks later.如果今天的花费大约是近期日均的两倍，Dollar 会标记出来。失控的循环、或选贵了的模型，当天就被抓住，而不是拖到三周后。
When one model dominates your spend, Dollar says so — e.g. "gpt-4o is your biggest cost; for simple tasks, gpt-4o-mini or Claude Haiku can cut spend." Cheaper where it makes sense, not everywhere.当某个模型占了大头，Dollar 会直说——比如"gpt-4o 是你最大的开销；简单任务用 gpt-4o-mini 或 Claude Haiku 能省下不少"。该省的地方省，而不是一刀切。
API spend is only half the story. Enter your flat subscriptions — ChatGPT Plus, Claude Pro — by hand, and Dollar folds those fixed fees into the monthly view so you see the whole cost, not just the metered part.API 花费只是故事的一半。把固定订阅——ChatGPT Plus、Claude Pro——手动填进去，Dollar 会把这些固定费用并入月度视图，让你看到完整成本，而不只是按量计费的那部分。
See it daily, take it anywhere — safely每日可见，随处可带——而且安全
Dollar shows you the day at a glance and lets you carry the numbers out clean, with personal paths and keys stripped on the way.Dollar 让你一眼看清这一天，也让你把数字干净地带走——导出途中自动抹去个人路径与密钥。
A daily chart, plus breakdowns by model and by project, show where spend builds up. Each morning Dollar greets you with a short report of yesterday's estimated cost — so the day starts with a number, not a guess.一张每日图表，加上按模型和按项目的拆解，让花费的来处一目了然。每天清晨，Dollar 用一份简短报告问候你，列出昨天的估算花费——让新的一天从一个数字开始，而不是一次猜测。
Export to 导出为 CSV · PDF · JSON for spreadsheets, a shareable report, or a full backup and restore. Every export auto-redacts your home path, username and API keys, so what leaves your machine carries the numbers but not your secrets.——用于表格分析、可分享的报告，或完整的备份与恢复。每一次导出都会自动抹去你的主目录路径、用户名和 API 密钥，离开这台机器的只有数字，没有你的秘密。
Be clear on this: except for OpenAI calls made through Dollar, every figure here is an estimate built from your local logs and an editable price table. It is a local-first guide to help you spend wisely — your real invoice comes from each provider, and actual billing may differ. Dollar never sells credits and never charges for usage; it is a one-time purchase on Steam.把话说清楚：除了经 Dollar 发出的 OpenAI 调用，这里的每一个数字都是据本地日志和一张可编辑价目表算出的估算。它是一份本地优先的参考，帮你把钱花得明白——真正的账单来自各家供应商，实际计费可能有出入。Dollar 从不出售额度，也从不按用量收费；它是 Steam 上的一次性买断。