给雇主
让候选人的 AI 助手直接读到贵公司系统里的岗位。免费,只读,十分钟。排序谁也买不到。
这是什么
OpenHire 是开源的求职工具(MCP server)。候选人把它装进 ChatGPT、Claude、豆包、DeepSeek(经支持 MCP 的客户端)、Cursor、腾讯 WorkBuddy 这类 AI 助手。
助手直接读雇主自己的招聘系统(Greenhouse、Lever、Ashby、北森、Moka、Workable、SmartRecruiters、Rippling、Personio)的公开接口。不经过招聘平台。候选人的简历只在他的电脑和贵公司的投递页之间,不经过我们。
每条岗位带贵公司系统里的原始发布日、在架天数和贵公司投递页的链接。数字的出处和口径在 numbers.json。
贵公司得到什么
岗位进候选人的助手
带贵公司系统的发布日,链接直达贵公司投递页。
认领后可以加说明
署贵公司名的说明,显示在指定岗位旁边。可标「常年开放」「难招但在招」「已停止招聘」。可更正日期含义(系统记的是需求创建日时)。可承诺回复时限。
谁也买不到什么
排序位置
排序只看匹配度和新鲜度。仓库里的测试锁死这一条。
更低的在架评分
评分只看重挂次数和在架天数。贵公司的说明显示在评分旁边,不改评分。
付费档位
没有。代码里没有一行计费。
贵公司在索引里吗
用 Greenhouse、Lever、Ashby、北森、Moka、Workable、SmartRecruiters、Rippling、Personio 的雇主,多半已经在。到仓库的雇主一节认领。
用飞书招聘、只在 BOSS 直聘发布、或用官网自建页面的雇主,现在不在。原因:
飞书的岗位接口要请求签名。我们把它当作访问控制,不绕过。
招聘平台不是贵公司自己的系统。我们不读。
自建页面多数没有机器可读的岗位数据。
用下面三条路的任何一条。都是只读、免费、十分钟。
三条路
① 发一张表(两分钟,不用 IT)
下载岗位表(Excel)。每行一个岗位,填六列:职位名称、工作城市、发布日期、投递链接或投递邮箱、薪资范围(可空)、岗位描述。填「公司信息」页。
用企业邮箱发到 gdchenhao@qq.com。主题写「岗位表 + 公司名」。正文写官网招聘页网址。3 个工作日内上线。不用营业执照。
每条岗位带「雇主自报」标记,在架天数从贵公司填的日期算。90 天后自动下架,还在招就再发一次表。投递链接只认贵公司网站上的 https 页面。
② 飞书招聘:授权两个只读权限(十分钟)
在飞书开放平台给我们的应用授权
hire:site:readonly和hire:site_job_post:readonly。只读贵公司已公开的招聘站和岗位。不读候选人数据。随时可以撤销。岗位带系统里的真实发布日,不用续期。
③ 官网放机器可读的岗位数据(要网站开发人员)
在招聘页加 schema.org
JobPosting的 JSON-LD(Google 招聘用的标准),或放一个静态 JSON 文件。把网址告诉我们。我们定期读页面。岗位更新时不用再发表。模板按需提供。
怎么联系
发表
把岗位表(两页都填)用企业邮箱发到 gdchenhao@qq.com。主题写「岗位表 + 公司名」。
GitHub 表单
填雇主认领(中文)。选「贵公司的岗位目前发布在哪里」,我们按那条路回。
没有 GitHub 账号
用企业邮箱发到 gdchenhao@qq.com。主题写「雇主认领 + 公司名」。
3 个工作日内给结论。身份材料只用于核验,不公开,不留存,不作他用。
我们怎么核实
企业邮箱只证明发件人控制这个域名。域名十块钱就能买,所以它只是第一步。我们还查这些:
工商登记
在国家企业信用信息公示系统查。公司存在,看成立日期、经营状态、经营范围。
ICP 备案
在工信部备案查询查。域名属于哪个法人。
域名年龄和网站历史
查注册日期和互联网档案馆。网站上线不满半年,不收。
回拨电话
我们打官网上公开的电话,不打邮件里的号码。
行业足迹
两个早于申请日的产品页、GitHub、论文或报道。证明公司在智驾、具身或 AI 基础设施里。
岗位本身
如果:
1. 境外高薪
2. 包机票
3. 不限经验
4. 客服、打字员这类骗局形态
拒收!
贵公司只填「公司信息」页:公司全称、统一社会信用代码(海外公司填注册号和注册地)、官网域名、ICP 备案号、官网公开电话所在页面和号码、两个行业足迹链接、经办人职务。核对由我们做。
我们核的是:公司存在,而且在我们的三个方向里。我们不替贵公司担保。每条岗位带「雇主自报」标记。一有举报,先下架。
边界
我们量得到岗位挂了多久,量不到贵公司想不想招。认领让贵公司用自己的话解释。
我们不在任何文案里点名「某家公司查不到」。走不走这三条路,由贵公司决定。
贵公司随时可以要求移出索引。我们照做。
For employers
Let candidates' AI assistants read the postings in your own hiring system. Free. Read-only. Ten minutes. Nobody can buy ranking.
What this is
OpenHire is an open-source job-search tool (an MCP server). Candidates add it to AI assistants such as ChatGPT, Claude, Doubao, DeepSeek (in an MCP client), Cursor and Tencent WorkBuddy.
The assistant reads the public API of the employer's own hiring system (Greenhouse, Lever, Ashby, Beisen, Moka, Workable, SmartRecruiters, Rippling, Personio). No job board sits in between. The candidate's résumé goes only between their computer and your apply page. It does not go through us.
Each posting has the original posting date from your system, days on the board, and a link to your apply page. The sources and counting rules of our numbers are in numbers.json.
What your company gets
Postings in candidates' assistants
With the posting date from your system and a direct link to your apply page.
Notes after you claim your listing
A note signed with your company name, next to the postings you choose. Labels: "always open", "hard to fill, still hiring", "no longer hiring". A date correction (when your system records the requisition date). An optional reply-time commitment.
What nobody can buy
Ranking position
Ranking uses match and freshness only. Tests in the repository lock this.
A lower time-on-board score
The score uses relists and days on the board only. Your note shows next to the score. It does not change the score.
Paid tiers
There are none. The code has no billing.
Is your company in the index?
If you use Greenhouse, Lever, Ashby, Beisen, Moka, Workable, SmartRecruiters, Rippling or Personio, you are probably in it. Claim your listing in the employers section of the repository.
If you use Feishu Hire, post only on BOSS Zhipin, or use a self-built careers page, you are not in it now. The reasons:
Feishu's job API needs signed requests. We treat that as access control and do not work around it.
A job board is not your own system. We do not read it.
Most self-built pages have no machine-readable job data.
Use any one of the three ways below. All are read-only, free and take ten minutes.
Three ways in
① Send a spreadsheet (two minutes, no IT)
Download the job sheet (Excel). Fill one row per posting, six columns: title, city, posting date, apply link or apply email, pay range (optional), description. Fill the "company information" sheet.
Send it from your company email to gdchenhao@qq.com. Subject: "Job sheet + company name". In the body, give your careers page address. Live within 3 working days. No business licence needed.
Each posting carries an "employer self-reported" label. Days on the board count from the date you give. Postings close after 90 days. If you are still hiring, send the sheet again. Apply links must be https pages on your own site.
② Feishu Hire: grant two read-only scopes (ten minutes)
On the Feishu open platform, grant our app
hire:site:readonlyandhire:site_job_post:readonly.These read only your public careers site and postings. No candidate data. You can revoke them at any time. Postings carry the real posting date. No renewal.
③ Put machine-readable job data on your site (needs a web developer)
Add schema.org
JobPostingJSON-LD to your careers page (the standard Google uses for jobs), or publish a static JSON file. Send us the address.We read the page on a schedule. No sheet to send when postings change. Templates on request.
How to reach us
Send the sheet
Send the job sheet (both sheets filled) from your company email to gdchenhao@qq.com. Subject: "Job sheet + company name".
GitHub form
Fill in Employer claim. Select where your postings are now. We reply on that path.
No GitHub account
Email gdchenhao@qq.com from your company domain. Subject: "Employer claim + company name".
Answer within 3 working days. We use identity documents only to verify. We do not publish, keep or reuse them.
How we verify
A company email proves only that the sender controls the domain. A domain costs a few dollars, so this is only the first step. We also check these:
Company register
National Enterprise Credit Information Publicity System: the company exists; founding date, status, scope.
ICP filing
MIIT filing lookup: which legal entity holds the domain.
Domain age and site history
Registration date and the Internet Archive. A site younger than six months is refused.
Call-back
We call the phone number on your public website, not the one in the email.
Industry footprint
Two product pages, GitHub repositories, papers or news items older than the application. They show the company works in autonomous driving, embodied AI or AI infrastructure.
The postings
If:
1. High pay abroad
2. Flights included
3. No experience required
4. Customer-service or typist scam patterns
Refused.
You fill in only the "company information" sheet: full legal name, unified social credit code (overseas: registration number and place), domain, ICP number, the page and number of your public phone, two footprint links, and the job title of the contact person. We do the checks.
We verify that the company exists and works in our three sectors. We do not vouch for your company. Every posting carries the "employer self-reported" label. On any report, we take postings down first.
Limits
We measure how long a posting has been up. We cannot measure whether you want to hire. A claim lets you explain in your own words.
We never name a company as "not readable" in any text to push it. Using these three ways is your decision.
You can ask us to remove your company from the index at any time. We do it.