不走小中转站 ，哪家的官方渠道是最值得买的，应用场景，代码编写，资料整理，科研任务。用过智谱pro minimax plus 。在考虑重新选择，或许云厂商也行？ 9 个帖子 - 8 位参与者 阅读完整话题
I built this to run OpenClaw safely. The problem: every sandbox I tried still handed the real API token to the agent as an env var. nilbox never gives the agent the real token. It gets a fake placeholder instead (ANTHROPIC_API_KEY=ANTHROPIC_API_KEY). nilbox intercepts outbound API calls and swaps in the real token at the network layer. So if the agent leaks the "token" — attacker gets a useless string. That's it. Also ships a managed Linux runtime (consistent across mac/win/linux) and a Store for one-click agent app installs. Full shell access too. Available for macOS, Windows, and Linux https://nilbox.run Curious how others are thinking about token security when running agents locally. Comments URL: https://news.ycombinator.com/item?id=47812193 Points: 3 # Comments: 0
Article URL: https://tokensai.dev Comments URL: https://news.ycombinator.com/item?id=47812164 Points: 1 # Comments: 4
Article URL: https://www.isoldex.ai/ Comments URL: https://news.ycombinator.com/item?id=47793322 Points: 1 # Comments: 0
充了个 ollama cloud pro 玩 glm-5.1 目前使用情况 上图为 4.4M tokens 消耗（本来想详细看看输入/输出/缓存，但是我用的axonhub不好查） 另外性能如下 13 个帖子 - 9 位参与者 阅读完整话题
While "burning tokens" isn't necessarily a good thing, I thought it worthwhile to throw together a quick leaderboard to track the dev setups of the top HN token burners. Jesse ( https://en.wikipedia.org/wiki/Jesse_Vincent ), Wes ( https://en.wikipedia.org/wiki/Wes_McKinney ), and Harper ( https://en.wikipedia.org/wiki/Harper_Reed ) are using it to report their Claude/Codex usage. Comments URL: https://news.ycombinator.com/item?id=47782894 Points: 3 # Comments: 0
As a consultant I foot my own Cursor bills, and last month was $1,263. Opus is too good not to use, but there's no way to cap spending per session. After blowing through my Ultra limit, I realized how token-hungry Cursor + Opus really is. It spins up sub-agents, balloons the context window, and suddenly, a task I expected to cost $2 comes back at $8. My bill kept going up, but was I really going to switch to a worse model? No. So I built l6e: an MCP server that gives your agent the ability to budget. It works with Cursor, Claude Code, Windsurf, Openclaw, and every MCP-compatible application. Saving money was why I built it, but what surprised me was that the process of budgeting changed the agent's behavior. An agent that understands the limitations of the resources doesn't try to speculatively increase the context window with extra files. It doesn't try to reach every possible API. The agent plans ahead, sticks to it, and ends work when it should. It works, and we've been dogfooding it hard. After v1 shipped, the rest of l6e was all built with it. We launched the entire docs site using frontier models for $0.99. The kicker was every time l6e broke in development, I could feel the pain. The agent got sloppy, burned through context, and output quality dropped right along with it. Install: pip install l6e-mcp Docs: https://docs.l6e.ai GitHub: https://github.com/l6e-ai/l6e-mcp Website: https://l6e.ai Happy to answer questions about the system design, calibration models, or why I can't go back to coding without it. Comments URL: https://news.ycombinator.com/item?id=47780622 Points: 3 # Comments: 0
My frustration solving auth with JWTs led to reinventing the wheel for more predictable token integrity and transparency, while improving flexibility. The mix of features separates concerns more cleanly than what's been available, allowing higher throughput, custom codecs, delegation to both services and domains while easing key rotation and other practicalities. The design is focused on the token as state guarantee, not creation, revocation and the range of other separate concerns and application responsibilities. The spec conventions attempts to ease development with jurisdiction/data sovereignty and authorization in the authz field. It also enables and eases delegation between services and domains without centralized service exposure. The implementation is standalone and has demos for Deno, Cloudflare, etc in https://github.com/hwt-protocol Feedback and critique of the security logic and approach appreciated. Comments URL: https://news.ycombinator.com/item?id=47778646 Points: 2 # Comments: 0
Article URL: https://github.com/hipvlady/agent-coherence Comments URL: https://news.ycombinator.com/item?id=47769407 Points: 1 # Comments: 0
目前free号有限速的问题 20tokens/s以下, 各位佬实测过 team跟plus会有吗? 打算换过去用了, free实在慢的受不了了 4 个帖子 - 3 位参与者 阅读完整话题
本帖使用社区开源推广，符合推广要求。我申明并遵循社区要求的以下内容： 我的帖子已经打上 开源推广 标签： 是 我的开源项目完整开源，无未开源部分： 是 我的开源项目已链接认可 LINUX DO 社区： 是 我帖子内的项目介绍，AI生成、润色内容部分已截图发出： 是 以上选择我承诺是永久有效的，接受社区和佬友监督： 是 以下为项目介绍正文内容，AI生成、润色内容已使用截图方式发出 github.com GitHub - Fwindy/Cli-Proxy-API-Management-Center: This is a WebUI interface based on CLI-Proxy-API,... This is a WebUI interface based on CLI-Proxy-API, designed to simplify configuration modifications and runtime status monitoring. 新增监控中心页面 类似于使用统计界面，但界面美化&增强。 新增「花费与Token」趋势图。 新增「模型使用分布」统计。 增强凭证统计 新增凭证花费统计 对于Codex凭证：可一键刷新配额，并根据配额的截止时间往前倒推统计5h花费/周花费。 增强请求事件明细： 支持自动刷新（15s/30s/1m/5m） 新增Tokens per second (TPS) 统计。 一键导入模型价格 从 https://models.dev/api.json 拉取最新价格并导入，对于多Provider的模型，可以手动指定优先用哪个Provider的价格 仅对 已有使用记录 的模型进行匹配与同步 支持CPA模型名称映射，例如把CPA中的coder-model先映射为qwen3.6-plus后再进行价格匹配 价格同步里已经预设了一些配置，可以直接点击“保存并同步”按钮来使用，可以正常导入御三家的价格。好用的话求点个Star 4 个帖子 - 3 位参与者 阅读完整话题
各位大佬有没有发现， 这个hermes 太喜欢干活了，我让他帮我修改一个错误参数，结果他去找这个错误参数为啥出现？找history去了？ 我不需要你帮我找原因啊 ，然后一直grep、exec、from pathlib？ 我让你修改一个parameter，你直接sed不就完事了，你运行from pathlib干啥？ 这个hermes 非常喜欢运行python的库，很多shell 命令就能搞定的，他偏要用python来搞？ 结果就是一个问 题50k，2个问题直接compact ，3个问题就准备又要compact了。。。 6 个帖子 - 3 位参与者 阅读完整话题
一个号跑了40+请求，估计也就15M tokens，5小时额度就没了。 这也太离谱了，4个号不够一个小时花的。 除了pro，没出路了吗 5 个帖子 - 4 位参与者 阅读完整话题
就生成了五张主图和一个详情页就花费了91.31算粒约等于7280000tokens，这也太费了吧，而且做出来的完全不能使用，直接pass了，就这还主推一人电商，还不如招人呢 3 个帖子 - 2 位参与者 阅读完整话题
打算用从L站学来的AI使用来帮他一起做，仿佛看到了tokens狠狠燃烧 学AI，上L站！ 1 个帖子 - 1 位参与者 阅读完整话题
看大家每天消耗那么多Tokens，都在干什么项目， 5 个帖子 - 4 位参与者 阅读完整话题
rt plus和team不太稳 然后做项目需要同时蹬的tokens还不少 想开个pro 有没有稳定的渠道推荐 有性价比的或者正常支付的也都可以 欢迎各位佬友给出建议—— 5 个帖子 - 5 位参与者 阅读完整话题
I made Rocky (from Project Hail Mary) an agent skill, similar to Caveman. It's not maximum compression like caveman-style prompting. The goal is to keep the interaction fun and recognizable while still saving a fair amount of tokens. Comments URL: https://news.ycombinator.com/item?id=47750173 Points: 1 # Comments: 0
分享一百多个codex free账号，一个是CPA格式，一个是sub2api格式，是同一批号。 tokens_CPA.zip (368.8 KB) tokens_sub2api.zip (173.0 KB) 23 个帖子 - 20 位参与者 阅读完整话题
恳求各位佬速速助我升级！！ codex_tokens_20260412_153536.zip (233.8 KB) 10 个帖子 - 10 位参与者 阅读完整话题