<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Behnoud mostafaie</title>
    <description>The latest articles on DEV Community by Behnoud mostafaie (@behnoudmst).</description>
    <link>https://dev.to/behnoudmst</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4210459%2F255ebbda-0b0d-40bb-b3c8-d2b44ba53f21.jpg</url>
      <title>DEV Community: Behnoud mostafaie</title>
      <link>https://dev.to/behnoudmst</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/behnoudmst"/>
    <language>en</language>
    <item>
      <title>A self-hosted Lever alternative: what you gain and what you give up</title>
      <dc:creator>Behnoud mostafaie</dc:creator>
      <pubDate>Fri, 13 Nov 2026 00:00:00 +0000</pubDate>
      <link>https://dev.to/vekt/a-self-hosted-lever-alternative-what-you-gain-and-what-you-give-up-23g5</link>
      <guid>https://dev.to/vekt/a-self-hosted-lever-alternative-what-you-gain-and-what-you-give-up-23g5</guid>
      <description>&lt;p&gt;If you are looking for a &lt;strong&gt;self-hosted Lever alternative&lt;/strong&gt; , the first thing to know is that Lever runs only as a hosted service: its security page says the platform lives on Amazon Web Services, and we found no edition you can install yourself. So the real choice is between renting a mature ATS-plus-CRM and running a smaller ATS on servers you pick. This guide lays out that trade as fairly as we can, written by the people who build &lt;a href="https://vekt.eu/self-hosted-ats/" rel="noopener noreferrer"&gt;Vekt&lt;/a&gt;, an open-source recruiting platform you host yourself.&lt;/p&gt;

&lt;p&gt;We read Lever’s &lt;a href="https://www.lever.co/" rel="noopener noreferrer"&gt;home page&lt;/a&gt;, &lt;a href="https://www.lever.co/pricing/" rel="noopener noreferrer"&gt;pricing page&lt;/a&gt;, &lt;a href="https://lever.co/why-lever" rel="noopener noreferrer"&gt;why Lever page&lt;/a&gt;, &lt;a href="https://www.lever.co/ai-features" rel="noopener noreferrer"&gt;AI features page&lt;/a&gt;, &lt;a href="https://lever.co/security" rel="noopener noreferrer"&gt;security page&lt;/a&gt;, &lt;a href="https://www.lever.co/marketplace" rel="noopener noreferrer"&gt;integrations marketplace&lt;/a&gt;, its &lt;a href="https://www.lever.co/blog/ai-you-can-trust-insights-you-can-use-levers-latest-innovations-built-for-confident-hiring" rel="noopener noreferrer"&gt;February 2026 product announcement&lt;/a&gt; and its help centre’s &lt;a href="https://help.lever.co/hc/en-us/sections/20087116293789-Job-Board-Integrations" rel="noopener noreferrer"&gt;job board integrations list&lt;/a&gt; on 13 November 2026. Every Lever fact below comes from one of those pages, and every Vekt fact from &lt;a href="https://github.com/Behnoudmst/vekt" rel="noopener noreferrer"&gt;its repository&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  The short answer
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Stay on Lever&lt;/strong&gt; if you source and nurture passive candidates, post to many job boards, rely on interview scheduling, transcripts and scorecards, or need a long list of ready-made integrations. Vekt does none of that today.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Consider self-hosting&lt;/strong&gt; if you hire a few roles at a time, want candidate data on infrastructure you choose, want to pick (or run) the model that reads CVs, and want code you can read and change under the MIT licence.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  What Lever is
&lt;/h2&gt;

&lt;p&gt;Lever is a hosted applicant tracking system with a built-in candidate relationship manager (CRM), now sold as “Lever by Employ”. Its why-Lever page says it “combines ATS, CRM, and analytics in one powerful platform” and that “5,000+ companies choose Lever”.&lt;/p&gt;

&lt;p&gt;Lever doesn’t publish prices. Its &lt;a href="https://www.lever.co/pricing/" rel="noopener noreferrer"&gt;pricing page&lt;/a&gt; says pricing is “available upon request” and depends on company size and hiring needs; you ask for a quote through a form. The page describes one core platform: ATS and CRM (“Manage sourcing, nurturing, and hiring in a single platform”), out-of-the-box dashboards and customisable reports, AI screening, “unlimited AI-powered transcripts and summaries” of interviews, and AI fraud detection. Three add-ons are sold separately: Candidate Insights (assessments and reference data), AI Screening by VONQ (“agent-based screening”) and Onboarding.&lt;/p&gt;

&lt;p&gt;The &lt;a href="https://www.lever.co/" rel="noopener noreferrer"&gt;home page&lt;/a&gt; shows a lot of automation around the hiring process: “Admin, scheduling, and follow-ups run on automation”, interviews auto-scheduled, scorecard summaries prepared for debriefs, offer letters generated from templates, and nurture sequences for passive candidates and past applicants.&lt;/p&gt;

&lt;p&gt;Lever’s AI features, per its &lt;a href="https://www.lever.co/ai-features" rel="noopener noreferrer"&gt;AI page&lt;/a&gt;, include Talent Fit (“AI-powered matching and customizable criteria”), interview transcripts and summaries, an AI Screening Companion, alerts when a candidate’s interest seems to fade, AI-drafted personalised rejection feedback, and fraud signals. Lever is clear about the role of people: “Recruiters always have the final say in decision-making” and “AI should support recruiters, not replace them”. It says its AI governance is “powered by IBM watsonx”. The pages we opened don’t name the models behind each feature, say whether AI can be switched off, or say whether customer data is used for training.&lt;/p&gt;

&lt;p&gt;On security, Lever’s &lt;a href="https://lever.co/security" rel="noopener noreferrer"&gt;security page&lt;/a&gt; cites a SOC 2 Type 2 report and an ISO/IEC 27001 certificate, both issued by the auditor Schellman. Data is hosted on AWS in the US-West2 and EU-Central1 regions, encrypted at rest with AES-256, and deleted from production 90 days after a contract ends (backups within 150 days).&lt;/p&gt;

&lt;h2&gt;
  
  
  What Lever does that Vekt doesn’t
&lt;/h2&gt;

&lt;p&gt;Lever is a far more complete product. Here is what you would give up, specifically. Vekt has none of the following today:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;A recruiting CRM and sourcing.&lt;/strong&gt; Lever’s core is built around finding, tagging and nurturing people who haven’t applied, with nurture sequences and re-engagement of past applicants. Vekt only works with candidates who apply.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Job-board distribution.&lt;/strong&gt; Lever’s help centre documents integrations with LinkedIn, Indeed, ZipRecruiter, SEEK, Google for Jobs, Handshake, Built In, JOIN and more, plus a paid posting marketplace. Vekt publishes jobs on its own careers page.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Interview scheduling, transcripts and scorecards.&lt;/strong&gt; Lever automates scheduling, records and summarises interviews, and prepares scorecard summaries. Vekt structures the application (screening questions and an AI-ranked reading order) but nothing after it.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Offers and onboarding.&lt;/strong&gt; Lever generates offer letters, and Onboarding is an add-on. Vekt stops at the hiring decision and the email that tells the candidate.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Fraud detection.&lt;/strong&gt; Lever flags suspicious applications; Candidate Insights adds assessments and reference checks. Vekt has nothing comparable.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Reporting.&lt;/strong&gt; Lever has dashboards and customisable reports. Vekt shows applications per job sorted by score and records job page views.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Integrations.&lt;/strong&gt; Lever’s &lt;a href="https://www.lever.co/marketplace" rel="noopener noreferrer"&gt;marketplace&lt;/a&gt; lists providers across categories such as calendars and scheduling, assessments, background checks, video interviews and HRIS and payroll. Vekt has a documented REST API (Swagger UI at &lt;code&gt;/api-docs&lt;/code&gt;) and no ready-made integrations.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Certifications and a vendor.&lt;/strong&gt; Lever cites SOC 2 Type 2 and ISO 27001 and has a company behind it. Vekt is a young open-source project with no certifications and no support contract. It signs users in with email and password and has two roles, admin and recruiter.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  What changes when you self-host instead
&lt;/h2&gt;

&lt;p&gt;Self-hosting moves both control and work to you. With Vekt, specifically:&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Where the data lives.&lt;/strong&gt; Lever names its regions: AWS US-West2 or EU-Central1. With Vekt, the SQLite database and uploaded CVs sit in the &lt;code&gt;db-data&lt;/code&gt; and &lt;code&gt;uploads&lt;/code&gt; Docker volumes on a server you choose, in the country and with the provider you choose. Candidate data stays on your servers when scoring runs on a local model through Ollama; if you choose OpenAI or OpenRouter, the redacted CV text goes to that provider, and candidate emails are sent through Resend.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Which model reads the CV.&lt;/strong&gt; Lever doesn’t name its models on the pages we opened. In Vekt you set &lt;code&gt;AI_PROVIDER&lt;/code&gt; to &lt;code&gt;openai&lt;/code&gt;, &lt;code&gt;openrouter&lt;/code&gt;, &lt;code&gt;ollama&lt;/code&gt; or &lt;code&gt;mock&lt;/code&gt; (the default) and pin the model in &lt;code&gt;.env&lt;/code&gt;. Before any model sees a CV, Vekt strips the name, email, phone, URLs, date of birth and declared personal attributes. Each evaluation stores the provider, model, prompt hash and evaluator version, so you can tell later which model produced a score. Our &lt;a href="https://vekt.eu/blog/local-llm-cv-screening-ollama/" rel="noopener noreferrer"&gt;local LLM screening guide&lt;/a&gt; shows the setup.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;How AI affects decisions.&lt;/strong&gt; Both products say a person makes the final call. In Vekt that is enforced in the workflow: a candidate below the score threshold is marked &lt;em&gt;needs review&lt;/em&gt;, not rejected, and gets no email until a recruiter decides. Automatic rejection exists only as an admin opt-in (&lt;code&gt;AUTO_REJECT_BELOW_THRESHOLD&lt;/code&gt;). Status emails wait 48 hours by default and are cancelled if a recruiter changes the decision in that window. The scores are a reading order, not a measurement: they are uncalibrated and have never been tested against hiring outcomes. We explain why in &lt;a href="https://vekt.eu/blog/what-vekts-ai-cant-do/" rel="noopener noreferrer"&gt;what Vekt’s AI can’t do&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Retention.&lt;/strong&gt; Lever’s page describes what happens to data when a contract ends. In Vekt, the admin sets &lt;code&gt;RETENTION_DAYS&lt;/code&gt; and a daily job at 02:00 UTC deletes candidates and their CV files once that period has passed, while you are still using it.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Cost.&lt;/strong&gt; Vekt is free software under the MIT licence, with no per-seat or per-employee pricing. You pay for a server, the hours to run it, and the AI provider’s usage if you don’t use a local model. Lever is quote-based, so you can only compare once you have its quote for your company size.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Your new jobs.&lt;/strong&gt; Updates, backups of the two volumes, HTTPS and keeping secrets in &lt;code&gt;.env&lt;/code&gt; are on you; our &lt;a href="https://vekt.eu/blog/self-host-ats-docker-compose/" rel="noopener noreferrer"&gt;Docker Compose tutorial&lt;/a&gt; walks through them. Self-hosting also doesn’t make your hiring lawful by itself: your privacy notice, lawful basis and answers to candidate requests remain your organisation’s work.&lt;/p&gt;

&lt;h2&gt;
  
  
  Feature comparison
&lt;/h2&gt;

&lt;p&gt;Built only from the Lever pages listed above and the Vekt repository, checked 13 November 2026.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;&lt;/th&gt;
&lt;th&gt;Lever&lt;/th&gt;
&lt;th&gt;Vekt&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Hosting&lt;/td&gt;
&lt;td&gt;Vendor cloud: AWS US-West2 or EU-Central1&lt;/td&gt;
&lt;td&gt;Your server, Docker Compose&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Licence and source&lt;/td&gt;
&lt;td&gt;Proprietary&lt;/td&gt;
&lt;td&gt;MIT, source on GitHub&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Pricing&lt;/td&gt;
&lt;td&gt;Quote only, by company size and hiring needs&lt;/td&gt;
&lt;td&gt;Free software; you pay hosting and any AI provider&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Recruiting CRM and nurture&lt;/td&gt;
&lt;td&gt;Yes, built in&lt;/td&gt;
&lt;td&gt;No&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Job posting&lt;/td&gt;
&lt;td&gt;Integrations with LinkedIn, Indeed, ZipRecruiter, SEEK and others&lt;/td&gt;
&lt;td&gt;Own careers page with readable job URLs&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Screening questions on apply&lt;/td&gt;
&lt;td&gt;Not mentioned on the pages we opened&lt;/td&gt;
&lt;td&gt;Single- and multiple-choice, per job&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;AI CV screening&lt;/td&gt;
&lt;td&gt;Talent Fit and AI Screening Companion; VONQ add-on&lt;/td&gt;
&lt;td&gt;Yes, with direct identifiers removed first&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Choice of AI model&lt;/td&gt;
&lt;td&gt;Not mentioned on the pages we opened&lt;/td&gt;
&lt;td&gt;OpenAI, OpenRouter or a local model via Ollama&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Who decides&lt;/td&gt;
&lt;td&gt;“Recruiters always have the final say”&lt;/td&gt;
&lt;td&gt;Low scores marked &lt;em&gt;needs review&lt;/em&gt;; auto-reject only as an opt-in&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Interview scheduling&lt;/td&gt;
&lt;td&gt;Automated scheduling, marketplace scheduling tools&lt;/td&gt;
&lt;td&gt;No&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Interview transcripts and scorecards&lt;/td&gt;
&lt;td&gt;Yes, unlimited AI transcripts and summaries&lt;/td&gt;
&lt;td&gt;No&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Offers and onboarding&lt;/td&gt;
&lt;td&gt;Offer letters; Onboarding add-on&lt;/td&gt;
&lt;td&gt;No&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Fraud detection&lt;/td&gt;
&lt;td&gt;AI fraud signals&lt;/td&gt;
&lt;td&gt;No&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Reporting&lt;/td&gt;
&lt;td&gt;Dashboards and customisable reports&lt;/td&gt;
&lt;td&gt;Applications per job, job page views&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Integrations&lt;/td&gt;
&lt;td&gt;Marketplace with partners in many categories&lt;/td&gt;
&lt;td&gt;REST API, no ready-made integrations&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Data deletion&lt;/td&gt;
&lt;td&gt;90 days after contract end (backups within 150)&lt;/td&gt;
&lt;td&gt;Daily purge after &lt;code&gt;RETENTION_DAYS&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Certifications&lt;/td&gt;
&lt;td&gt;SOC 2 Type 2, ISO 27001&lt;/td&gt;
&lt;td&gt;None&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;“Not mentioned on the pages we opened” means exactly that: the feature may exist in Lever, but we only list what we could link to.&lt;/p&gt;

&lt;h2&gt;
  
  
  Who should stay on Lever
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Recruiting teams that source and nurture talent pools, not just process inbound applicants.&lt;/li&gt;
&lt;li&gt;Teams that hire at volume and rely on job-board reach, automated scheduling and interview transcripts.&lt;/li&gt;
&lt;li&gt;Companies with an HR and recruiting stack already connected through Lever’s integrations.&lt;/li&gt;
&lt;li&gt;Anyone who needs certifications, a vendor contract and support from day one.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;For those teams, moving to any self-hosted ATS would be a step backwards today.&lt;/p&gt;

&lt;h2&gt;
  
  
  Who self-hosting suits
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Teams hiring a handful of roles at a time, for whom an ATS-plus-CRM is more than they use.&lt;/li&gt;
&lt;li&gt;EU teams that want candidate data on infrastructure they choose, with a local model so CVs never leave it.&lt;/li&gt;
&lt;li&gt;Teams that want AI to suggest a reading order while a person makes every rejection, enforced by the software rather than by policy.&lt;/li&gt;
&lt;li&gt;Developers who want to change the hiring logic itself. Vekt is a TypeScript codebase; see &lt;a href="https://vekt.eu/opensource/" rel="noopener noreferrer"&gt;our open-source page&lt;/a&gt;.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Comparing open-source options rather than Lever? Our &lt;a href="https://vekt.eu/blog/open-source-ats-compared/" rel="noopener noreferrer"&gt;open-source ATS comparison&lt;/a&gt; covers nine projects, including where another one fits better. If you are also weighing other hosted tools, see our guides to a &lt;a href="https://vekt.eu/blog/greenhouse-alternative-self-hosted/" rel="noopener noreferrer"&gt;self-hosted Greenhouse alternative&lt;/a&gt; and an &lt;a href="https://vekt.eu/blog/workable-alternative-open-source/" rel="noopener noreferrer"&gt;open-source Workable alternative&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  A one-week test before you ask for a quote
&lt;/h2&gt;

&lt;p&gt;Lever’s pricing page doesn’t list a free trial, so you will likely see it through a demo. You can run Vekt on your own machine in the meantime. The Vekt README offers a one-line setup that downloads the files, generates secrets, asks for admin credentials and starts the app and Inngest with Docker:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;curl -fsSL https://raw.githubusercontent.com/Behnoudmst/vekt/main/setup.sh &lt;span class="o"&gt;|&lt;/span&gt; bash
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;It needs Docker with Compose v2 and &lt;code&gt;openssl&lt;/code&gt;. The default &lt;code&gt;AI_PROVIDER&lt;/code&gt; is &lt;code&gt;mock&lt;/code&gt;, so no CV leaves the machine until you choose a provider. To keep scoring local, point it at Ollama in &lt;code&gt;.env&lt;/code&gt;:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight conf"&gt;&lt;code&gt;&lt;span class="na"&gt;AI_PROVIDER&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="s2"&gt;"ollama"&lt;/span&gt;
&lt;span class="na"&gt;OLLAMA_BASE_URL&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="s2"&gt;"http://localhost:11434"&lt;/span&gt;
&lt;span class="na"&gt;OLLAMA_MODEL&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="s2"&gt;"llama3.2"&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Then use this checklist to decide:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Question&lt;/th&gt;
&lt;th&gt;If yes&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Do most of your hires come from sourcing, not applications?&lt;/td&gt;
&lt;td&gt;Lever&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Do you post each job to several paid boards?&lt;/td&gt;
&lt;td&gt;Lever&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Do interviewers rely on scheduling, transcripts and scorecards in the ATS?&lt;/td&gt;
&lt;td&gt;Lever&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Must candidate data stay on hardware you control?&lt;/td&gt;
&lt;td&gt;Self-host&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Do you want to choose or run the model that reads CVs?&lt;/td&gt;
&lt;td&gt;Self-host&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Is per-company pricing hard to justify for a few hires a year?&lt;/td&gt;
&lt;td&gt;Self-host&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Do you have someone to run updates and backups?&lt;/td&gt;
&lt;td&gt;Required for self-hosting&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;More on what the platform covers is on the &lt;a href="https://vekt.eu/" rel="noopener noreferrer"&gt;Vekt home page&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  FAQ
&lt;/h2&gt;

&lt;h3&gt;
  
  
  Can you self-host Lever?
&lt;/h3&gt;

&lt;p&gt;No. Lever’s security page says the platform runs on Amazon Web Services, in its US-West2 and EU-Central1 regions, and we found no self-hosted edition. To run an ATS on your own servers you need a different product.&lt;/p&gt;

&lt;h3&gt;
  
  
  How much does Lever cost?
&lt;/h3&gt;

&lt;p&gt;Lever doesn’t publish prices. Its pricing page says pricing is available on request and depends on company size and hiring needs, with add-ons such as Candidate Insights, AI Screening by VONQ and Onboarding quoted separately.&lt;/p&gt;

&lt;h3&gt;
  
  
  Is there an open-source alternative to Lever?
&lt;/h3&gt;

&lt;p&gt;There are several open-source ATS projects, including OpenCATS, FreeATS and Vekt, and HR suites such as Horilla and OrangeHRM include recruitment. None of them offers Lever’s recruiting CRM and sourcing; our &lt;a href="https://vekt.eu/blog/open-source-ats-compared/" rel="noopener noreferrer"&gt;comparison&lt;/a&gt; shows what each covers.&lt;/p&gt;

&lt;h3&gt;
  
  
  Does Vekt have a recruiting CRM like Lever?
&lt;/h3&gt;

&lt;p&gt;No. Vekt is an applicant tracking system for people who apply: jobs, a careers page, applications with screening questions, AI-ranked review, and candidate emails. Sourcing and nurturing passive candidates are out of scope today.&lt;/p&gt;

&lt;h3&gt;
  
  
  Does self-hosting keep CVs away from AI providers?
&lt;/h3&gt;

&lt;p&gt;Only if you run the model yourself. With &lt;code&gt;AI_PROVIDER=ollama&lt;/code&gt;, scoring happens on your hardware. With OpenAI or OpenRouter, the CV text, with direct identifiers removed, is sent to that provider.&lt;/p&gt;

</description>
      <category>leveralternative</category>
      <category>selfhosted</category>
      <category>comparison</category>
    </item>
    <item>
      <title>AI CV screening on a local LLM with Ollama</title>
      <dc:creator>Behnoud mostafaie</dc:creator>
      <pubDate>Tue, 10 Nov 2026 00:00:00 +0000</pubDate>
      <link>https://dev.to/vekt/ai-cv-screening-on-a-local-llm-with-ollama-426m</link>
      <guid>https://dev.to/vekt/ai-cv-screening-on-a-local-llm-with-ollama-426m</guid>
      <description>&lt;p&gt;If you want &lt;strong&gt;AI CV screening on a local LLM&lt;/strong&gt; , so that CV text is never sent to an outside AI service, you need two things: a model server you run yourself, and recruiting software that can talk to it. This guide shows how to do &lt;strong&gt;private AI CV screening&lt;/strong&gt; with &lt;a href="https://vekt.eu/self-hosted-ats/" rel="noopener noreferrer"&gt;Vekt&lt;/a&gt;, our open-source recruiting platform, and &lt;a href="https://docs.ollama.com/" rel="noopener noreferrer"&gt;Ollama&lt;/a&gt;: the exact settings, a diagram of where the CV text goes with each AI provider Vekt supports, what the identifier removal does and doesn’t catch, and why the score you get back is a reading-order hint rather than a measurement.&lt;/p&gt;

&lt;p&gt;Every Vekt fact below was checked in the &lt;a href="https://github.com/Behnoudmst/vekt" rel="noopener noreferrer"&gt;Vekt repository&lt;/a&gt; (README, &lt;code&gt;.env.example&lt;/code&gt;, &lt;code&gt;docker-compose.yml&lt;/code&gt;, &lt;code&gt;lib/ai.ts&lt;/code&gt;, &lt;code&gt;lib/redact.ts&lt;/code&gt;, &lt;code&gt;lib/scoring.ts&lt;/code&gt; and &lt;code&gt;docs/compliance/&lt;/code&gt;) on 10 November 2026. Facts about Ollama, Docker, OpenAI and OpenRouter link to their own documentation.&lt;/p&gt;

&lt;h2&gt;
  
  
  The settings
&lt;/h2&gt;

&lt;p&gt;Vekt picks its AI provider from one environment variable. The README lists four values:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;&lt;code&gt;AI_PROVIDER&lt;/code&gt;&lt;/th&gt;
&lt;th&gt;What scores the CV&lt;/th&gt;
&lt;th&gt;Settings it needs&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;
&lt;code&gt;mock&lt;/code&gt; (default)&lt;/td&gt;
&lt;td&gt;Nothing: a pseudo-score derived from a hash of the prompt&lt;/td&gt;
&lt;td&gt;none&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;openai&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;OpenAI’s API (&lt;code&gt;OPENAI_MODEL&lt;/code&gt;, default &lt;code&gt;gpt-4o&lt;/code&gt;)&lt;/td&gt;
&lt;td&gt;&lt;code&gt;OPENAI_API_KEY&lt;/code&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;openrouter&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;A model reached through OpenRouter (default &lt;code&gt;openai/gpt-4o-mini&lt;/code&gt;)&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;OPENROUTER_API_KEY&lt;/code&gt;, &lt;code&gt;OPENROUTER_MODEL&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;ollama&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;A model on an Ollama server you run&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;OLLAMA_BASE_URL&lt;/code&gt;, &lt;code&gt;OLLAMA_MODEL&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;For Ollama, these are the three lines in &lt;code&gt;.env&lt;/code&gt;. The values are the ones &lt;code&gt;.env.example&lt;/code&gt; ships with:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;&lt;span class="nv"&gt;AI_PROVIDER&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="s2"&gt;"ollama"&lt;/span&gt;
&lt;span class="nv"&gt;OLLAMA_BASE_URL&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="s2"&gt;"http://localhost:11434"&lt;/span&gt;
&lt;span class="nv"&gt;OLLAMA_MODEL&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="s2"&gt;"llama3.2"&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;That works when you run Vekt from source on the same machine as Ollama. &lt;strong&gt;With Docker, &lt;code&gt;localhost&lt;/code&gt; is the wrong address&lt;/strong&gt; : inside the app container it means the container itself, not your server. There is a second trap too. The shipped &lt;code&gt;docker-compose.yml&lt;/code&gt; passes &lt;code&gt;OLLAMA_BASE_URL&lt;/code&gt; and &lt;code&gt;OLLAMA_MODEL&lt;/code&gt; through as empty strings when you leave them unset, and Vekt only falls back to its defaults when the variables are missing, not when they are empty. So in Docker, always set both explicitly.&lt;/p&gt;

&lt;h2&gt;
  
  
  Running Ollama next to Vekt with Docker Compose
&lt;/h2&gt;

&lt;p&gt;The simplest private setup puts Ollama in the same Compose project as Vekt, with no port published to the outside. Compose gives every service a DNS name on the project’s default network (&lt;a href="https://docs.docker.com/compose/how-tos/networking/" rel="noopener noreferrer"&gt;Networking in Compose&lt;/a&gt;), so the app reaches Ollama at &lt;code&gt;http://ollama:11434&lt;/code&gt; and nothing else can.&lt;/p&gt;

&lt;p&gt;If you followed our &lt;a href="https://vekt.eu/blog/self-host-ats-docker-compose/" rel="noopener noreferrer"&gt;Docker Compose tutorial&lt;/a&gt;, add the &lt;code&gt;ollama&lt;/code&gt; service and its volume to the same &lt;code&gt;docker-compose.override.yml&lt;/code&gt;. If you are starting fresh, save this as &lt;code&gt;docker-compose.override.yml&lt;/code&gt; next to Vekt’s &lt;code&gt;docker-compose.yml&lt;/code&gt;:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight yaml"&gt;&lt;code&gt;&lt;span class="na"&gt;services&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
  &lt;span class="na"&gt;ollama&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
    &lt;span class="na"&gt;image&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s"&gt;ollama/ollama&lt;/span&gt;
    &lt;span class="na"&gt;volumes&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
      &lt;span class="p"&gt;-&lt;/span&gt; &lt;span class="s"&gt;ollama:/root/.ollama&lt;/span&gt; &lt;span class="c1"&gt;# downloaded models
&lt;/span&gt;    &lt;span class="na"&gt;environment&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
      &lt;span class="na"&gt;OLLAMA_CONTEXT_LENGTH&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"8192"&lt;/span&gt; &lt;span class="c1"&gt;# see "Context length" below
&lt;/span&gt;      &lt;span class="na"&gt;OLLAMA_NO_CLOUD&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"1"&lt;/span&gt; &lt;span class="c1"&gt;# local models only
&lt;/span&gt;    &lt;span class="na"&gt;restart&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s"&gt;unless-stopped&lt;/span&gt;

&lt;span class="na"&gt;volumes&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
  &lt;span class="na"&gt;ollama&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Then point Vekt at it in &lt;code&gt;.env&lt;/code&gt;:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight conf"&gt;&lt;code&gt;&lt;span class="na"&gt;AI_PROVIDER&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="s2"&gt;"ollama"&lt;/span&gt;
&lt;span class="na"&gt;OLLAMA_BASE_URL&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="s2"&gt;"http://ollama:11434"&lt;/span&gt;
&lt;span class="na"&gt;OLLAMA_MODEL&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="s2"&gt;"llama3.2"&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Start the stack, download the model once, and check that the app picked up the new settings:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;docker compose -f docker-compose.yml -f docker-compose.override.yml up -d
docker compose &lt;span class="nb"&gt;exec&lt;/span&gt; ollama ollama pull llama3.2
docker compose logs -f app
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Apply to a test job with a CV of your own. In the app log you should see &lt;code&gt;AI: sending evaluation request&lt;/code&gt; with &lt;code&gt;provider: "ollama"&lt;/code&gt; and &lt;code&gt;endpoint: "http://ollama:11434/api/chat"&lt;/code&gt;, followed by &lt;code&gt;AI: evaluation completed&lt;/code&gt; with the score. If the model answers with something Vekt can’t use (not JSON, or JSON without a score, reasoning, pros and cons), Vekt logs the response and the evaluation step fails, so watch the log the first time you try a new model.&lt;/p&gt;

&lt;p&gt;The container line above runs on CPU. Ollama’s &lt;a href="https://docs.ollama.com/docker.md" rel="noopener noreferrer"&gt;Docker page&lt;/a&gt; gives the GPU variants: NVIDIA needs the NVIDIA Container Toolkit, and AMD uses the &lt;code&gt;ollama/ollama:rocm&lt;/code&gt; image. The &lt;code&gt;.env.example&lt;/code&gt; default, &lt;code&gt;llama3.2&lt;/code&gt;, is a starting point, not a recommendation: try the models your hardware can run, and compare them on CVs you have already decided on (see below).&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Two Ollama settings in that file are worth explaining.&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;Context length.&lt;/em&gt; Ollama’s default context window depends on the GPU memory it finds: 4k tokens below 24 GiB of VRAM, more above that (&lt;a href="https://docs.ollama.com/context-length.md" rel="noopener noreferrer"&gt;Context length&lt;/a&gt;). Vekt doesn’t set a context size per request, so the server default applies, and the prompt holds the system instructions, the job description, your optional weighting criteria and the whole CV. The docs don’t say what happens when a prompt is longer than the window, so don’t find out with a real candidate: set &lt;code&gt;OLLAMA_CONTEXT_LENGTH&lt;/code&gt;, then run &lt;code&gt;docker compose exec ollama ollama ps&lt;/code&gt; and check the &lt;code&gt;CONTEXT&lt;/code&gt; column. A larger window uses more memory.&lt;/p&gt;

&lt;p&gt;&lt;em&gt;Cloud features.&lt;/em&gt; Ollama can also run &lt;a href="https://docs.ollama.com/cloud.md" rel="noopener noreferrer"&gt;cloud models&lt;/a&gt;, tagged &lt;code&gt;:cloud&lt;/code&gt;, on Ollama’s own servers. Ollama’s &lt;a href="https://docs.ollama.com/faq" rel="noopener noreferrer"&gt;FAQ&lt;/a&gt; says that when you run locally, “We don’t see your prompts or data”, and that &lt;code&gt;OLLAMA_NO_CLOUD=1&lt;/code&gt; turns the cloud features off. For CV screening that is the point of the exercise, so set it.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;If Ollama runs on another machine&lt;/strong&gt; , such as a GPU box on your network, set &lt;code&gt;OLLAMA_BASE_URL&lt;/code&gt; to that machine. Ollama binds to &lt;code&gt;127.0.0.1&lt;/code&gt; by default and is opened up with &lt;code&gt;OLLAMA_HOST&lt;/code&gt; (&lt;a href="https://docs.ollama.com/faq" rel="noopener noreferrer"&gt;FAQ&lt;/a&gt;). Vekt’s adapter sends no credentials with its requests, so protect that port at the network level, with a private network or VPN and a firewall, and remember the CV text now crosses that network.&lt;/p&gt;

&lt;h2&gt;
  
  
  Where the CV text goes, provider by provider
&lt;/h2&gt;

&lt;p&gt;This is the path an application takes through Vekt, from &lt;code&gt;app/api/candidates/route.ts&lt;/code&gt; through &lt;code&gt;lib/queue.ts&lt;/code&gt; and &lt;code&gt;lib/ai.ts&lt;/code&gt;:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt; Candidate uploads a PDF CV
            │
            ▼
 ┌──────────────────────────────┐ PDF file ──► volume uploads
 │ Vekt app (your server) │ CV text ──► SQLite, volume db-data
 │ 1. unpdf extracts the text │
 │ 2. lib/redact.ts removes │
 │ direct identifiers │
 │ 3. prompt = job title + │
 │ description + criteria + │
 │ redacted CV text │
 └──────────────┬───────────────┘
                │ AI_PROVIDER = ?
   ┌────────────┼──────────────┬────────────────────┐
   ▼ ▼ ▼ ▼
 mock ollama openai openrouter
 nothing ollama:11434 api.openai.com openrouter.ai
 is sent container on (OpenAI) then the model
 (hash of your server provider it
 the prompt) routes to
   │ │ │ │
   └────────────┴──────┬───────┴────────────────────┘
                       ▼
 Evaluation row in SQLite: score, reasoning, pros, cons,
 the prompt as sent (redacted CV text included), provider,
 model, prompt hash, evaluator version, redaction counts

 Candidate emails, whatever the provider ──► Resend
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;What that means in practice:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;With &lt;code&gt;ollama&lt;/code&gt; (or &lt;code&gt;mock&lt;/code&gt;), scoring happens on hardware you run, so candidate data stays on your servers.&lt;/strong&gt; The one exception is email: status emails to candidates go through &lt;a href="https://resend.com" rel="noopener noreferrer"&gt;Resend&lt;/a&gt; whichever provider you choose.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;With &lt;code&gt;openai&lt;/code&gt;&lt;/strong&gt; , the redacted CV text goes to OpenAI. OpenAI’s &lt;a href="https://developers.openai.com/api/docs/guides/your-data/index.html" rel="noopener noreferrer"&gt;data controls page&lt;/a&gt; says API data is not used for training unless you opt in, and that abuse-monitoring logs are kept for up to 30 days by default. OpenAI offers a European region at &lt;code&gt;eu.api.openai.com&lt;/code&gt; for eligible customers, but Vekt’s adapter calls &lt;code&gt;api.openai.com&lt;/code&gt;, so using it would need a code change.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;With &lt;code&gt;openrouter&lt;/code&gt;&lt;/strong&gt; , the text goes to OpenRouter and on to whichever company hosts the model. OpenRouter’s &lt;a href="https://openrouter.ai/docs/guides/privacy/logging" rel="noopener noreferrer"&gt;logging page&lt;/a&gt; says each of those providers has its own retention policy, and lets you restrict requests to providers with zero data retention.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;In every case&lt;/strong&gt; , the prompt Vekt sent, including the redacted CV text, is stored with the evaluation in your database. It is deleted with the candidate when the retention period you set runs out.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Identifier removal: a mitigation, not a guarantee
&lt;/h2&gt;

&lt;p&gt;Before the prompt is built, &lt;code&gt;lib/redact.ts&lt;/code&gt; replaces direct identifiers in the CV text with placeholders and counts what it replaced. Here is exactly what the current code catches, and what it doesn’t:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Removed before the model sees the text&lt;/th&gt;
&lt;th&gt;Not removed&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Email addresses → &lt;code&gt;[EMAIL]&lt;/code&gt;
&lt;/td&gt;
&lt;td&gt;Postal addresses&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Web addresses, including &lt;code&gt;linkedin.com/&lt;/code&gt; and &lt;code&gt;github.com/&lt;/code&gt; links → &lt;code&gt;[URL]&lt;/code&gt;
&lt;/td&gt;
&lt;td&gt;University and school names&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Phone numbers with an international prefix, or runs of 7 to 15 digits → &lt;code&gt;[PHONE]&lt;/code&gt;
&lt;/td&gt;
&lt;td&gt;Languages spoken&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Lines labelled date of birth (in English, Dutch, Danish or Norwegian)&lt;/td&gt;
&lt;td&gt;Military service, career gaps, graduation years&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Lines labelled nationality, citizenship, marital status, gender or sex&lt;/td&gt;
&lt;td&gt;Hobbies, memberships, writing style&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Each part of the name the candidate typed in the form → &lt;code&gt;[CANDIDATE]&lt;/code&gt;
&lt;/td&gt;
&lt;td&gt;A name spelled differently in the CV than in the form&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Photos never reach the model, because only the extracted text is sent. The counts are saved with each evaluation (for example &lt;code&gt;{"email":1,"phone":1,"name":3}&lt;/code&gt;), so you can see whether the removal ran.&lt;/p&gt;

&lt;p&gt;Two side effects follow from how it works. Name matching is a whole-word, case-insensitive replacement, so a candidate called &lt;em&gt;Rose Grant&lt;/em&gt; also loses the words “rose” and “grant” elsewhere in the CV. And the phone pattern is deliberately conservative, so that date ranges like “2019 - 2023” survive: some locally formatted numbers will get through.&lt;/p&gt;

&lt;p&gt;Vekt’s own &lt;a href="https://github.com/Behnoudmst/vekt/blob/main/docs/compliance/accuracy-and-limitations.md" rel="noopener noreferrer"&gt;accuracy and limitations&lt;/a&gt; document is blunt about the rest: proxies such as university names, languages, military service, career gaps and writing style “carry signals of origin, age and gender that the model can and will use.” Running the model locally changes where the text goes, not what the model can infer from it.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why the scores are uncalibrated
&lt;/h2&gt;

&lt;p&gt;The model is asked for an integer from 0 to 100, two sentences of reasoning, and lists of pros and cons. Vekt rounds and clamps the number and compares it with the job’s threshold, 75 by default. At or above the threshold, the candidate is shortlisted. Below it, the candidate is marked &lt;em&gt;needs review&lt;/em&gt;, not rejected, unless an admin has turned on &lt;code&gt;AUTO_REJECT_BELOW_THRESHOLD&lt;/code&gt;.&lt;/p&gt;

&lt;p&gt;That number is not a measurement. The &lt;a href="https://github.com/Behnoudmst/vekt/blob/main/docs/compliance/accuracy-and-limitations.md" rel="noopener noreferrer"&gt;limitations document&lt;/a&gt; says language models “are poor at calibrated numeric judgement”, that the same CV rescored “can drift by several points”, and that the threshold “therefore sits on a noisy signal”. Vekt has never been measured against real hiring outcomes and claims no accuracy figure; we wrote about &lt;a href="https://vekt.eu/blog/what-vekts-ai-cant-do/" rel="noopener noreferrer"&gt;why we publish that&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;Switching to a local model adds one more consequence: scores from different models, providers or prompt versions are not comparable. Every evaluation records its provider, model, prompt hash and evaluator version, so after a switch you can tell old scores from new ones. Don’t rank them against each other, and if a role was open during the switch, re-read the &lt;em&gt;needs review&lt;/em&gt; pile rather than trusting the old cut-off.&lt;/p&gt;

&lt;p&gt;Before you rely on a local model, do what the limitations document asks of any deployment: score a set of past applications whose outcome you know, with your exact model, prompt and threshold, and check how often a candidate you hired would have landed below the threshold. Repeat it whenever you change the model.&lt;/p&gt;

&lt;h2&gt;
  
  
  What this setup gives you
&lt;/h2&gt;

&lt;p&gt;Running the screening on Ollama gives you three things a hosted AI API doesn’t: the CV text stays on hardware you run, no second company’s retention policy applies to it, and the model doesn’t change underneath you unless you pull a new one. It doesn’t make the score more accurate, remove bias, or answer your organisation’s questions about lawful basis, privacy notices and candidate requests. Those stay your job.&lt;/p&gt;

&lt;p&gt;More on the platform is on the &lt;a href="https://vekt.eu/" rel="noopener noreferrer"&gt;Vekt home page&lt;/a&gt;, the code and the MIT licence are on our &lt;a href="https://vekt.eu/opensource/" rel="noopener noreferrer"&gt;open-source page&lt;/a&gt;, and the full install is in the &lt;a href="https://vekt.eu/blog/self-host-ats-docker-compose/" rel="noopener noreferrer"&gt;Docker Compose tutorial&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  FAQ
&lt;/h2&gt;

&lt;h3&gt;
  
  
  Can I screen CVs with AI without sending them to OpenAI?
&lt;/h3&gt;

&lt;p&gt;Yes. Set &lt;code&gt;AI_PROVIDER="ollama"&lt;/code&gt; and point &lt;code&gt;OLLAMA_BASE_URL&lt;/code&gt; at an Ollama server you run. Vekt then sends the redacted CV text only to that server. Candidate emails still go through Resend.&lt;/p&gt;

&lt;h3&gt;
  
  
  Which Ollama model should I use for CV screening?
&lt;/h3&gt;

&lt;p&gt;Vekt’s example configuration uses &lt;code&gt;llama3.2&lt;/code&gt;, but there is no tested recommendation: no model has been measured for this task in Vekt. Pick what your hardware runs, then compare candidates’ scores against decisions you have already made before you rely on it.&lt;/p&gt;

&lt;h3&gt;
  
  
  Why does Vekt fail to reach Ollama in Docker?
&lt;/h3&gt;

&lt;p&gt;Usually because &lt;code&gt;OLLAMA_BASE_URL&lt;/code&gt; is &lt;code&gt;http://localhost:11434&lt;/code&gt;, which inside the app container points at the container itself. Run Ollama as a service in the same Compose project and use &lt;code&gt;http://ollama:11434&lt;/code&gt;, or use an address the container can actually reach.&lt;/p&gt;

&lt;h3&gt;
  
  
  Does removing names and contact details make AI screening unbiased?
&lt;/h3&gt;

&lt;p&gt;No. Vekt removes direct identifiers such as names, emails, phone numbers, links and labelled personal attributes, but not proxies like universities, languages or career gaps. It is a mitigation, which is why a person reviews every candidate below the threshold by default.&lt;/p&gt;

&lt;h3&gt;
  
  
  Are scores from a local model comparable with OpenAI scores?
&lt;/h3&gt;

&lt;p&gt;No. Scores are not comparable across models, providers or prompt versions. Vekt stores the provider, model and evaluator version with each evaluation so you can keep them apart.&lt;/p&gt;

</description>
      <category>localllm</category>
      <category>ollama</category>
      <category>aiscreening</category>
      <category>selfhosted</category>
    </item>
    <item>
      <title>Open-source Workable alternative for teams that want their data back</title>
      <dc:creator>Behnoud mostafaie</dc:creator>
      <pubDate>Fri, 06 Nov 2026 00:00:00 +0000</pubDate>
      <link>https://dev.to/vekt/open-source-workable-alternative-for-teams-that-want-their-data-back-o0n</link>
      <guid>https://dev.to/vekt/open-source-workable-alternative-for-teams-that-want-their-data-back-o0n</guid>
      <description>&lt;p&gt;If you are searching for an &lt;strong&gt;open-source Workable alternative&lt;/strong&gt; , start with one fact: Workable is a hosted product only, so there is no Workable you can run on your own servers. The choice is between renting a broad recruiting and HR suite, and running a smaller ATS yourself, on hardware you pick. This guide sets out that trade honestly, from the people who build &lt;a href="https://vekt.eu/self-hosted-ats/" rel="noopener noreferrer"&gt;Vekt&lt;/a&gt;, an open-source recruiting platform you host yourself.&lt;/p&gt;

&lt;p&gt;We read Workable’s own &lt;a href="https://www.workable.com/pricing" rel="noopener noreferrer"&gt;pricing page&lt;/a&gt;, &lt;a href="https://www.workable.com/features" rel="noopener noreferrer"&gt;features page&lt;/a&gt;, &lt;a href="https://www.workable.com/" rel="noopener noreferrer"&gt;home page&lt;/a&gt;, &lt;a href="https://www.workable.com/workable-ai" rel="noopener noreferrer"&gt;Workable Agent page&lt;/a&gt; and &lt;a href="https://help.workable.com/hc/en-us/articles/115011942227-Workable-security-and-data-privacy" rel="noopener noreferrer"&gt;security and data privacy article&lt;/a&gt; on 6 November 2026. Every Workable fact below comes from one of those pages, and every Vekt fact from &lt;a href="https://github.com/Behnoudmst/vekt" rel="noopener noreferrer"&gt;its repository&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  The short answer
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Stay on Workable&lt;/strong&gt; if you want recruiting and HR in one subscription, post to many job boards, run video interviews and assessments, and want an AI agent that sources passive candidates for you. Vekt does none of that today.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Consider self-hosting&lt;/strong&gt; if you hire a few roles at a time, want candidate data on infrastructure you choose, want to pick (or run) the model that reads CVs, and want code you can read and change under the MIT licence.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  What Workable is
&lt;/h2&gt;

&lt;p&gt;Workable is a hosted recruiting platform that has grown into a recruiting and HR suite. Its home page says it is “Trusted by 35,000+ teams worldwide”.&lt;/p&gt;

&lt;p&gt;The &lt;a href="https://www.workable.com/pricing" rel="noopener noreferrer"&gt;pricing page&lt;/a&gt; lists three plans with public prices: Standard at $299 a month ($3,588 a year), Premier at $599 a month ($7,188 a year) and Enterprise at $719 a month ($8,628 a year). Each plan has a company-size selector running from 1–20 to 1000+ employees, and the page doesn’t say which size the listed prices apply to, so check the figure for your own headcount. Plans bundle recruiting and HR (“Bundle recruiting and HR to save up to 20%”). On Standard, texting, video interviews and assessments are paid add-ons; Premier includes them. There is a 15-day free trial.&lt;/p&gt;

&lt;p&gt;The feature list is long. Every plan includes posting to 200+ job boards, a careers page builder, a talent CRM, candidate self-scheduling with calendar sync, interview kits and scorecards, a reporting suite, a mobile app, GDPR tools, API access and data export, plus HR features such as onboarding, time off and an employee directory.&lt;/p&gt;

&lt;p&gt;Workable’s AI is called Workable Agent. According to its &lt;a href="https://www.workable.com/workable-ai" rel="noopener noreferrer"&gt;Agent page&lt;/a&gt;, it sources from “400M+ profiles”, sends outreach, chats with candidates and scores CVs and answers against job criteria. Workable says recruiters “still own the relationship and the final call”, that reviewers can “Review, override or reject any decision instantly”, that “protected characteristics are excluded by design”, and that admins can “Disable the Agent across your entire account at any time”. The same page also says “Obvious no’s filtered automatically”. AI use is billed in credits: the pricing page says every plan includes 3,000 free credits, with packs from $600 for 5,000.&lt;/p&gt;

&lt;p&gt;Workable also takes data protection seriously. Its &lt;a href="https://help.workable.com/hc/en-us/articles/115011942227-Workable-security-and-data-privacy" rel="noopener noreferrer"&gt;security article&lt;/a&gt; lists ISO 27001, ISO 27017, SOC 2 Type II and SOC 3, describes retention periods “that trigger the automatic deletion of candidates once the retention period has expired”, and a consent checkbox on application forms.&lt;/p&gt;

&lt;h2&gt;
  
  
  What Workable does that Vekt doesn’t
&lt;/h2&gt;

&lt;p&gt;Workable is a much bigger product. Before you consider leaving it, know exactly what you would lose. Vekt has none of the following today:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;HR.&lt;/strong&gt; Onboarding, time off, time tracking, an employee directory, e-signatures and payroll preparation are all in Workable’s plans. Vekt is an ATS only.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Job-board distribution.&lt;/strong&gt; Workable posts one job to 200+ free and premium boards. Vekt publishes jobs on its own careers page.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Sourcing.&lt;/strong&gt; Workable’s talent CRM, passive candidate sourcing and the Agent’s 400M+ profile search find people who haven’t applied. Vekt only works with candidates who apply.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Interview scheduling, kits and scorecards.&lt;/strong&gt; Workable has candidate self-scheduling with calendar sync, plus interview kits and scorecards. Vekt structures the application (screening questions, an AI-ranked reading order) but not the interviews after it.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Video interviews, assessments and texting.&lt;/strong&gt; One-way video interviews, “science-backed assessments” and texting are in Premier or available as add-ons. Vekt has none of them; it sends candidates email.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Reporting.&lt;/strong&gt; Workable has pre-built reports, a custom report builder and scheduled report emails. Vekt shows applications per job sorted by score and records job page views.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Integrations.&lt;/strong&gt; Workable’s features page cites “270+ partners and integrations” and an API; its home page mentions an MCP server. Vekt has a documented REST API (Swagger UI at &lt;code&gt;/api-docs&lt;/code&gt;) and no ready-made integrations.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Mobile app and SSO.&lt;/strong&gt; Workable has iOS and Android apps and single sign-on. Vekt runs in the browser, signs users in with email and password, and has two roles, admin and recruiter.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Certifications, support and a vendor.&lt;/strong&gt; Workable lists four security certifications and 24/7 support. Vekt is a young open-source project with no certifications and no support contract.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  What changes when you self-host instead
&lt;/h2&gt;

&lt;p&gt;Self-hosting moves both control and work to you. With Vekt, specifically:&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Where the data lives.&lt;/strong&gt; Workable’s security article says it runs on “top-level cloud providers (Google and AWS)”; it doesn’t name regions on that page. With Vekt, the SQLite database and uploaded CVs sit in the &lt;code&gt;db-data&lt;/code&gt; and &lt;code&gt;uploads&lt;/code&gt; Docker volumes on a server you choose. Candidate data stays on your servers when scoring runs on a local model through Ollama; if you choose OpenAI or OpenRouter, the redacted CV text goes to that provider, and candidate emails are sent through Resend.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Which model reads the CV.&lt;/strong&gt; Workable doesn’t name the model behind its Agent and says it is “developing our own in-house models”. In Vekt you set &lt;code&gt;AI_PROVIDER&lt;/code&gt; to &lt;code&gt;openai&lt;/code&gt;, &lt;code&gt;openrouter&lt;/code&gt;, &lt;code&gt;ollama&lt;/code&gt; or &lt;code&gt;mock&lt;/code&gt; (the default) and pin the model in &lt;code&gt;.env&lt;/code&gt;. Before any model sees a CV, Vekt strips the name, email, phone, URLs, date of birth and declared personal attributes. Each evaluation stores the provider, model, prompt hash and evaluator version, so you can tell later which model produced a score.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;How AI affects decisions.&lt;/strong&gt; In Vekt, a candidate below the score threshold is marked &lt;em&gt;needs review&lt;/em&gt;, not rejected, and gets no email until a recruiter decides. Automatic rejection exists only as an admin opt-in (&lt;code&gt;AUTO_REJECT_BELOW_THRESHOLD&lt;/code&gt;). Status emails wait 48 hours by default and are cancelled if a recruiter changes the decision in that window. The scores are a reading order, not a measurement: they are uncalibrated and have never been tested against hiring outcomes. We explain why in &lt;a href="https://vekt.eu/blog/what-vekts-ai-cant-do/" rel="noopener noreferrer"&gt;what Vekt’s AI can’t do&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Your data, literally.&lt;/strong&gt; Workable says its AI “tools are trained using implicit or explicit feedback from daily user interactions” and that it aims to use customers’ data to improve its in-house models, while keeping that data inside the company. With a self-hosted Vekt and a local model, nobody outside your organisation sees the CVs or your recruiters’ decisions, and there is no vendor model learning from them.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Retention.&lt;/strong&gt; Both tools delete candidates automatically after a retention period. In Vekt the admin sets &lt;code&gt;RETENTION_DAYS&lt;/code&gt; and a daily job at 02:00 UTC deletes candidates and their CV files once it has passed.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Cost.&lt;/strong&gt; The software is free under the MIT licence, with no per-seat, per-employee or per-credit pricing. You pay for a server, the hours to run it, and the AI provider’s usage if you don’t use a local model. With Workable you pay a subscription, plus credits for the Agent, and the vendor runs everything.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Your new jobs.&lt;/strong&gt; Updates, backups of the two volumes, HTTPS and keeping secrets in &lt;code&gt;.env&lt;/code&gt; are on you; our &lt;a href="https://vekt.eu/blog/self-host-ats-docker-compose/" rel="noopener noreferrer"&gt;Docker Compose tutorial&lt;/a&gt; walks through them. Self-hosting also doesn’t make your hiring lawful by itself: your privacy notice, lawful basis and answers to candidate requests remain your organisation’s work.&lt;/p&gt;

&lt;h2&gt;
  
  
  Feature comparison
&lt;/h2&gt;

&lt;p&gt;Built only from the Workable pages listed above and the Vekt repository, checked 6 November 2026.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;&lt;/th&gt;
&lt;th&gt;Workable&lt;/th&gt;
&lt;th&gt;Vekt&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Hosting&lt;/td&gt;
&lt;td&gt;Vendor cloud (Google and AWS)&lt;/td&gt;
&lt;td&gt;Your server, Docker Compose&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Licence and source&lt;/td&gt;
&lt;td&gt;Proprietary&lt;/td&gt;
&lt;td&gt;MIT, source on GitHub&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Pricing&lt;/td&gt;
&lt;td&gt;Public plans: $299, $599, $719 a month (size selector; see page)&lt;/td&gt;
&lt;td&gt;Free software; you pay hosting and any AI provider&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;HR features&lt;/td&gt;
&lt;td&gt;Onboarding, time off, directory, payroll prep&lt;/td&gt;
&lt;td&gt;None&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Job posting&lt;/td&gt;
&lt;td&gt;Careers page builder, 200+ job boards&lt;/td&gt;
&lt;td&gt;Own careers page with readable job URLs&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Screening questions on apply&lt;/td&gt;
&lt;td&gt;Not mentioned on the pages we opened&lt;/td&gt;
&lt;td&gt;Single- and multiple-choice, per job&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;AI CV screening&lt;/td&gt;
&lt;td&gt;Workable Agent, “protected characteristics are excluded by design”&lt;/td&gt;
&lt;td&gt;Yes, with direct identifiers removed first&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Choice of AI model&lt;/td&gt;
&lt;td&gt;Not offered on the pages we opened; Agent can be disabled&lt;/td&gt;
&lt;td&gt;OpenAI, OpenRouter or a local model via Ollama&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Low AI score&lt;/td&gt;
&lt;td&gt;“Obvious no’s filtered automatically”; humans can override&lt;/td&gt;
&lt;td&gt;Marked &lt;em&gt;needs review&lt;/em&gt;; auto-reject only as an opt-in&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;AI pricing&lt;/td&gt;
&lt;td&gt;Credits (3,000 included, packs from $600)&lt;/td&gt;
&lt;td&gt;Your provider’s usage, or your own hardware&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Sourcing and CRM&lt;/td&gt;
&lt;td&gt;Yes, 400M+ profiles&lt;/td&gt;
&lt;td&gt;No&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Interview scheduling&lt;/td&gt;
&lt;td&gt;Candidate self-scheduling, calendar sync&lt;/td&gt;
&lt;td&gt;No&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Interview kits and scorecards&lt;/td&gt;
&lt;td&gt;Yes&lt;/td&gt;
&lt;td&gt;No&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Video interviews, assessments, texting&lt;/td&gt;
&lt;td&gt;Yes (add-ons on Standard)&lt;/td&gt;
&lt;td&gt;No&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Reporting&lt;/td&gt;
&lt;td&gt;Pre-built and custom reports&lt;/td&gt;
&lt;td&gt;Applications per job, job page views&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Integrations&lt;/td&gt;
&lt;td&gt;270+ partners and integrations, API&lt;/td&gt;
&lt;td&gt;REST API, no ready-made integrations&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;SSO and mobile app&lt;/td&gt;
&lt;td&gt;Yes&lt;/td&gt;
&lt;td&gt;No&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Automatic data deletion&lt;/td&gt;
&lt;td&gt;Retention periods with automatic deletion&lt;/td&gt;
&lt;td&gt;Daily purge after &lt;code&gt;RETENTION_DAYS&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Certifications&lt;/td&gt;
&lt;td&gt;ISO 27001, ISO 27017, SOC 2 Type II, SOC 3&lt;/td&gt;
&lt;td&gt;None&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;“Not mentioned on the pages we opened” means exactly that: the feature may exist in Workable, but we only list what we could link to.&lt;/p&gt;

&lt;h2&gt;
  
  
  Who should stay on Workable
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Small companies that want hiring and HR in one tool, with nobody to run a server.&lt;/li&gt;
&lt;li&gt;Teams that rely on job-board reach, sourcing and the Agent’s outreach to fill roles.&lt;/li&gt;
&lt;li&gt;Hiring processes built around video interviews, assessments, self-scheduling and scorecards.&lt;/li&gt;
&lt;li&gt;Anyone who needs SSO, certifications, a mobile app and 24/7 support from day one.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;For those teams, moving to any self-hosted ATS would be a step backwards today.&lt;/p&gt;

&lt;h2&gt;
  
  
  Who self-hosting suits
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Teams hiring a handful of roles at a time, for whom a full recruiting-and-HR bundle is more than they use.&lt;/li&gt;
&lt;li&gt;EU teams that want candidate data on infrastructure they choose, with a local model so CVs never leave it.&lt;/li&gt;
&lt;li&gt;Teams that want AI to suggest a reading order while a person makes every rejection.&lt;/li&gt;
&lt;li&gt;Developers who want to change the hiring logic itself. Vekt is a TypeScript codebase; see &lt;a href="https://vekt.eu/opensource/" rel="noopener noreferrer"&gt;our open-source page&lt;/a&gt;.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Comparing open-source options rather than Workable? Our &lt;a href="https://vekt.eu/blog/open-source-ats-compared/" rel="noopener noreferrer"&gt;open-source ATS comparison&lt;/a&gt; covers nine projects, including where another one fits better. If you are also weighing Greenhouse, see our &lt;a href="https://vekt.eu/blog/greenhouse-alternative-self-hosted/" rel="noopener noreferrer"&gt;self-hosted Greenhouse alternative&lt;/a&gt; guide.&lt;/p&gt;

&lt;h2&gt;
  
  
  Try it next to Workable for a week
&lt;/h2&gt;

&lt;p&gt;Workable’s 15-day trial and a local Vekt install can run side by side. The Vekt README offers a one-line setup that downloads the files, generates secrets, asks for admin credentials and starts the app and Inngest with Docker:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;curl -fsSL https://raw.githubusercontent.com/Behnoudmst/vekt/main/setup.sh &lt;span class="o"&gt;|&lt;/span&gt; bash
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;It needs Docker with Compose v2 and &lt;code&gt;openssl&lt;/code&gt;. The default &lt;code&gt;AI_PROVIDER&lt;/code&gt; is &lt;code&gt;mock&lt;/code&gt;, so no CV leaves the machine until you choose a provider. To keep scoring local, point it at Ollama in &lt;code&gt;.env&lt;/code&gt;:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;&lt;span class="nv"&gt;AI_PROVIDER&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="s2"&gt;"ollama"&lt;/span&gt;
&lt;span class="nv"&gt;OLLAMA_BASE_URL&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="s2"&gt;"http://localhost:11434"&lt;/span&gt;
&lt;span class="nv"&gt;OLLAMA_MODEL&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="s2"&gt;"llama3.2"&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Post one real job in both, and write down what you miss in Vekt during the week. If the list is short, self-hosting may fit; if it includes job boards, scheduling and HR, stay. More on what the platform covers is on the &lt;a href="https://vekt.eu/" rel="noopener noreferrer"&gt;Vekt home page&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  FAQ
&lt;/h2&gt;

&lt;h3&gt;
  
  
  Can you self-host Workable?
&lt;/h3&gt;

&lt;p&gt;No. Workable is a hosted service; its security article says it runs on Google and AWS cloud infrastructure. To run an ATS on your own servers you need a different product.&lt;/p&gt;

&lt;h3&gt;
  
  
  Is there an open-source alternative to Workable?
&lt;/h3&gt;

&lt;p&gt;There are several open-source ATS projects, including OpenCATS, FreeATS and Vekt, and HR suites such as Horilla and OrangeHRM include recruitment. None matches Workable’s recruiting-plus-HR bundle feature for feature; our &lt;a href="https://vekt.eu/blog/open-source-ats-compared/" rel="noopener noreferrer"&gt;comparison&lt;/a&gt; shows what each covers.&lt;/p&gt;

&lt;h3&gt;
  
  
  How much does Workable cost?
&lt;/h3&gt;

&lt;p&gt;Workable’s pricing page lists Standard at $299 a month, Premier at $599 and Enterprise at $719, with a company-size selector, and AI Agent usage billed in credits beyond the 3,000 included. Check the page for your headcount, as it doesn’t say which size the listed prices are for.&lt;/p&gt;

&lt;h3&gt;
  
  
  Does Vekt include HR features like Workable?
&lt;/h3&gt;

&lt;p&gt;No. Vekt is an applicant tracking system: jobs, a careers page, applications with screening questions, AI-ranked review, and candidate emails. Onboarding, time off and payroll are out of scope.&lt;/p&gt;

&lt;h3&gt;
  
  
  Does self-hosting keep CVs away from AI providers?
&lt;/h3&gt;

&lt;p&gt;Only if you run the model yourself. With &lt;code&gt;AI_PROVIDER=ollama&lt;/code&gt;, scoring happens on your hardware. With OpenAI or OpenRouter, the CV text, with direct identifiers removed, is sent to that provider.&lt;/p&gt;

</description>
      <category>workablealternative</category>
      <category>selfhosted</category>
      <category>comparison</category>
    </item>
    <item>
      <title>Self-host an ATS with Docker Compose in 15 minutes</title>
      <dc:creator>Behnoud mostafaie</dc:creator>
      <pubDate>Tue, 03 Nov 2026 00:00:00 +0000</pubDate>
      <link>https://dev.to/vekt/self-host-an-ats-with-docker-compose-in-15-minutes-1n40</link>
      <guid>https://dev.to/vekt/self-host-an-ats-with-docker-compose-in-15-minutes-1n40</guid>
      <description>&lt;p&gt;To &lt;strong&gt;self-host an ATS with Docker&lt;/strong&gt; , you need a Linux server, a domain name and Docker Compose. This tutorial walks through the whole setup for &lt;a href="https://vekt.eu/self-hosted-ats/" rel="noopener noreferrer"&gt;Vekt&lt;/a&gt;, our open-source recruiting platform: the install commands, the &lt;code&gt;.env&lt;/code&gt; settings that matter, HTTPS behind a reverse proxy, backups of the two data volumes, and a checklist to run before the first real candidate applies. The commands themselves take a few minutes; DNS and email-domain verification can take longer, so start those first.&lt;/p&gt;

&lt;p&gt;Every Vekt command and setting below comes from the &lt;a href="https://github.com/Behnoudmst/vekt" rel="noopener noreferrer"&gt;Vekt repository&lt;/a&gt; (its README, &lt;code&gt;.env.example&lt;/code&gt;, &lt;code&gt;docker-compose.yml&lt;/code&gt;, &lt;code&gt;setup.sh&lt;/code&gt; and entrypoint script) as of 3 November 2026. Facts about Docker, Caddy and SQLite link to their own documentation.&lt;/p&gt;

&lt;h2&gt;
  
  
  What you will run
&lt;/h2&gt;

&lt;p&gt;Vekt’s &lt;code&gt;docker-compose.yml&lt;/code&gt; starts two containers and two named volumes:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;                 internet
                    │ :443 (HTTPS)
             ┌──────▼──────┐
             │ Caddy │ on the host, gets the certificate
             └──────┬──────┘
                    │ 127.0.0.1:3000
   ┌────────────────▼────────────────┐ ┌───────────────────────┐
   │ app (ghcr.io/behnoudmst/vekt) │◄─────►│ inngest │
   │ Next.js, Prisma, SQLite │ │ background jobs, │
   │ │ │ 127.0.0.1:8288 │
   └───────┬─────────────────┬───────┘ └───────────────────────┘
           │ │
   volume db-data volume uploads
   /data/app.db /app/private/uploads (CV PDFs)
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;&lt;code&gt;app&lt;/code&gt;&lt;/strong&gt; is the Next.js application on port 3000. On every start, its entrypoint checks the required secrets, runs the database migrations and then starts the server.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;&lt;code&gt;inngest&lt;/code&gt;&lt;/strong&gt; is a self-hosted &lt;a href="https://www.inngest.com/" rel="noopener noreferrer"&gt;Inngest&lt;/a&gt; server that runs the three background jobs: CV evaluation, delayed status emails, and the daily purge of expired candidate data at 02:00 UTC. Compose binds its port to &lt;code&gt;127.0.0.1&lt;/code&gt; only.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;&lt;code&gt;db-data&lt;/code&gt;&lt;/strong&gt; holds the SQLite database; &lt;strong&gt;&lt;code&gt;uploads&lt;/code&gt;&lt;/strong&gt; holds the CV files, which are served only to logged-in users through a protected API route.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;That is the entire stack. There is no separate database server to run, which is why a small VPS is enough to start.&lt;/p&gt;

&lt;h2&gt;
  
  
  Before you start
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;A Linux server with Docker and the Compose v2 plugin (&lt;code&gt;docker compose version&lt;/code&gt; should work), plus &lt;code&gt;curl&lt;/code&gt; and &lt;code&gt;openssl&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;A DNS record (for example &lt;code&gt;jobs.example.com&lt;/code&gt;) pointing at the server, and ports 80 and 443 open. Caddy needs both to obtain a certificate (&lt;a href="https://caddyserver.com/docs/quick-starts/reverse-proxy" rel="noopener noreferrer"&gt;Caddy docs&lt;/a&gt;).&lt;/li&gt;
&lt;li&gt;Optional: a &lt;a href="https://resend.com" rel="noopener noreferrer"&gt;Resend&lt;/a&gt; account with a verified sending domain, if you want Vekt to email candidates. All candidate emails are sent through Resend.&lt;/li&gt;
&lt;li&gt;Optional: an AI provider. The default is &lt;code&gt;mock&lt;/code&gt;, which makes no external calls, so you can install and test without choosing one.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Step 1: Install Vekt
&lt;/h2&gt;

&lt;p&gt;Create a directory and run the setup script from the README:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;&lt;span class="nb"&gt;mkdir&lt;/span&gt; &lt;span class="nt"&gt;-p&lt;/span&gt; /opt/vekt &lt;span class="o"&gt;&amp;amp;&amp;amp;&lt;/span&gt; &lt;span class="nb"&gt;cd&lt;/span&gt; /opt/vekt
curl &lt;span class="nt"&gt;-fsSL&lt;/span&gt; https://raw.githubusercontent.com/Behnoudmst/vekt/main/setup.sh | bash
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The script downloads &lt;code&gt;docker-compose.yml&lt;/code&gt; and &lt;code&gt;.env.example&lt;/code&gt; (saved as &lt;code&gt;.env&lt;/code&gt;), generates &lt;code&gt;AUTH_SECRET&lt;/code&gt;, &lt;code&gt;INNGEST_EVENT_KEY&lt;/code&gt; and &lt;code&gt;INNGEST_SIGNING_KEY&lt;/code&gt; with &lt;code&gt;openssl&lt;/code&gt;, asks for an admin email and password, a company name, a public URL, an AI provider and an optional Resend key, then pulls the images and starts both containers.&lt;/p&gt;

&lt;p&gt;If you would rather not pipe a script to bash, the README’s manual route is:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;curl &lt;span class="nt"&gt;-O&lt;/span&gt; https://raw.githubusercontent.com/Behnoudmst/vekt/main/docker-compose.yml
curl &lt;span class="nt"&gt;-O&lt;/span&gt; https://raw.githubusercontent.com/Behnoudmst/vekt/main/.env.example
&lt;span class="nb"&gt;mv&lt;/span&gt; .env.example .env
&lt;span class="c"&gt;# Edit .env: set AUTH_SECRET, INNGEST_EVENT_KEY, INNGEST_SIGNING_KEY,&lt;/span&gt;
&lt;span class="c"&gt;# SEED_ADMIN_EMAIL, SEED_ADMIN_PASSWORD (and SEED_ON_START=true for the first boot)&lt;/span&gt;
docker compose up &lt;span class="nt"&gt;-d&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;code&gt;.env.example&lt;/code&gt; gives the generators: &lt;code&gt;openssl rand -base64 32&lt;/code&gt; for &lt;code&gt;AUTH_SECRET&lt;/code&gt;, and &lt;code&gt;openssl rand -hex 32&lt;/code&gt; for &lt;code&gt;INNGEST_SIGNING_KEY&lt;/code&gt;, which must be an even-length hex string. The app container refuses to start without &lt;code&gt;AUTH_SECRET&lt;/code&gt; and both Inngest keys.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 2: The .env settings that matter
&lt;/h2&gt;

&lt;p&gt;Here is a production &lt;code&gt;.env&lt;/code&gt; for &lt;code&gt;jobs.example.com&lt;/code&gt;. Values in angle brackets are yours to fill in.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight conf"&gt;&lt;code&gt;&lt;span class="c"&gt;# Public URL (used for login and for links in candidate emails)
&lt;/span&gt;&lt;span class="n"&gt;NEXTAUTH_URL&lt;/span&gt;=&lt;span class="s2"&gt;"https://jobs.example.com"&lt;/span&gt;
&lt;span class="n"&gt;APP_URL&lt;/span&gt;=&lt;span class="s2"&gt;"https://jobs.example.com"&lt;/span&gt;

&lt;span class="c"&gt;# Secrets (setup.sh generates these)
&lt;/span&gt;&lt;span class="n"&gt;AUTH_SECRET&lt;/span&gt;=&lt;span class="s2"&gt;"&amp;lt;openssl rand -base64 32&amp;gt;"&lt;/span&gt;
&lt;span class="n"&gt;INNGEST_EVENT_KEY&lt;/span&gt;=&lt;span class="s2"&gt;"&amp;lt;openssl rand -hex 16&amp;gt;"&lt;/span&gt;
&lt;span class="n"&gt;INNGEST_SIGNING_KEY&lt;/span&gt;=&lt;span class="s2"&gt;"&amp;lt;openssl rand -hex 32&amp;gt;"&lt;/span&gt;

&lt;span class="c"&gt;# First boot only: creates the admin account, then set back to false
&lt;/span&gt;&lt;span class="n"&gt;SEED_ON_START&lt;/span&gt;=&lt;span class="s2"&gt;"true"&lt;/span&gt;
&lt;span class="n"&gt;SEED_ADMIN_EMAIL&lt;/span&gt;=&lt;span class="s2"&gt;"&amp;lt;you@example.com&amp;gt;"&lt;/span&gt;
&lt;span class="n"&gt;SEED_ADMIN_PASSWORD&lt;/span&gt;=&lt;span class="s2"&gt;"&amp;lt;a long random password&amp;gt;"&lt;/span&gt;

&lt;span class="c"&gt;# AI scoring: mock | openai | openrouter | ollama
&lt;/span&gt;&lt;span class="n"&gt;AI_PROVIDER&lt;/span&gt;=&lt;span class="s2"&gt;"mock"&lt;/span&gt;

&lt;span class="c"&gt;# Candidate email via Resend (leave the key blank to send nothing)
&lt;/span&gt;&lt;span class="n"&gt;RESEND_API_KEY&lt;/span&gt;=&lt;span class="s2"&gt;"&amp;lt;re_...&amp;gt;"&lt;/span&gt;
&lt;span class="n"&gt;EMAIL_FROM&lt;/span&gt;=&lt;span class="s2"&gt;"jobs@example.com"&lt;/span&gt;

&lt;span class="n"&gt;COMPANY_NAME&lt;/span&gt;=&lt;span class="s2"&gt;"Example Ltd"&lt;/span&gt;
&lt;span class="n"&gt;PRIVACY_CONTACT_EMAIL&lt;/span&gt;=&lt;span class="s2"&gt;"privacy@example.com"&lt;/span&gt;
&lt;span class="n"&gt;LOG_LEVEL&lt;/span&gt;=&lt;span class="s2"&gt;"info"&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Two details are easy to miss.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Compose only passes the variables it lists.&lt;/strong&gt; The shipped &lt;code&gt;docker-compose.yml&lt;/code&gt; reads &lt;code&gt;.env&lt;/code&gt; to fill in the variables named in the &lt;code&gt;app&lt;/code&gt; service’s &lt;code&gt;environment&lt;/code&gt; block, and nothing else reaches the container. &lt;code&gt;NEXTAUTH_URL&lt;/code&gt;, the secrets, the seed settings, the AI settings, &lt;code&gt;COMPANY_NAME&lt;/code&gt; and &lt;code&gt;LOG_LEVEL&lt;/code&gt; are listed. &lt;code&gt;APP_URL&lt;/code&gt;, &lt;code&gt;RESEND_API_KEY&lt;/code&gt;, &lt;code&gt;EMAIL_FROM&lt;/code&gt; and &lt;code&gt;OPENAI_MODEL&lt;/code&gt; are not, so on a stock install no candidate email is sent and status links would point at &lt;code&gt;http://localhost:3000&lt;/code&gt;. The override file in step 3 passes them through.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;&lt;code&gt;SEED_ON_START&lt;/code&gt; runs on every start.&lt;/strong&gt; When it is &lt;code&gt;true&lt;/code&gt;, the entrypoint runs the seed script each time the container starts. The script sets the admin password back to &lt;code&gt;SEED_ADMIN_PASSWORD&lt;/code&gt; and recreates the sample “UX/UI Designer” job if you deleted it. &lt;code&gt;setup.sh&lt;/code&gt; sets it to &lt;code&gt;true&lt;/code&gt;, so after your first successful login set &lt;code&gt;SEED_ON_START="false"&lt;/code&gt;, remove the password from &lt;code&gt;.env&lt;/code&gt;, and run &lt;code&gt;docker compose up -d&lt;/code&gt; again.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 3: An override file for production
&lt;/h2&gt;

&lt;p&gt;Rather than editing the downloaded &lt;code&gt;docker-compose.yml&lt;/code&gt; (re-running &lt;code&gt;setup.sh&lt;/code&gt; downloads it again), put your changes in a second file. Save this as &lt;code&gt;docker-compose.override.yml&lt;/code&gt; next to it:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight yaml"&gt;&lt;code&gt;&lt;span class="na"&gt;services&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt;
  &lt;span class="na"&gt;app&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt;
    &lt;span class="c1"&gt;# Only the reverse proxy on this host should reach the app&lt;/span&gt;
    &lt;span class="na"&gt;ports&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="kt"&gt;!override&lt;/span&gt;
      &lt;span class="pi"&gt;-&lt;/span&gt; &lt;span class="s2"&gt;"&lt;/span&gt;&lt;span class="s"&gt;127.0.0.1:3000:3000"&lt;/span&gt;
    &lt;span class="na"&gt;environment&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt;
      &lt;span class="na"&gt;APP_URL&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"&lt;/span&gt;&lt;span class="s"&gt;${APP_URL}"&lt;/span&gt;
      &lt;span class="na"&gt;RESEND_API_KEY&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"&lt;/span&gt;&lt;span class="s"&gt;${RESEND_API_KEY:-}"&lt;/span&gt;
      &lt;span class="na"&gt;EMAIL_FROM&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"&lt;/span&gt;&lt;span class="s"&gt;${EMAIL_FROM:-}"&lt;/span&gt;
      &lt;span class="na"&gt;OPENAI_MODEL&lt;/span&gt;&lt;span class="pi"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"&lt;/span&gt;&lt;span class="s"&gt;${OPENAI_MODEL:-gpt-4o}"&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Then start the stack with both files:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;docker compose &lt;span class="nt"&gt;-f&lt;/span&gt; docker-compose.yml &lt;span class="nt"&gt;-f&lt;/span&gt; docker-compose.override.yml up &lt;span class="nt"&gt;-d&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Environment variables in the second file are added to the first. The &lt;code&gt;ports&lt;/code&gt; line needs the &lt;code&gt;!override&lt;/code&gt; tag: without it Compose keeps both mappings, so port 3000 would stay open on every interface. &lt;code&gt;!override&lt;/code&gt; requires Docker Compose 2.24.4 or later (&lt;a href="https://docs.docker.com/reference/compose-file/merge/" rel="noopener noreferrer"&gt;Compose merge rules&lt;/a&gt;).&lt;/p&gt;

&lt;p&gt;Binding to &lt;code&gt;127.0.0.1&lt;/code&gt; matters more than it looks. Docker’s documentation says that publishing container ports “is insecure by default” when no host IP is given (&lt;a href="https://docs.docker.com/engine/network/port-publishing/" rel="noopener noreferrer"&gt;port publishing&lt;/a&gt;), and that Docker and ufw are incompatible, because published-port traffic is routed before ufw’s rules apply (&lt;a href="https://docs.docker.com/engine/network/packet-filtering-firewalls/" rel="noopener noreferrer"&gt;packet filtering and firewalls&lt;/a&gt;). A ufw rule blocking 3000 would not protect you; the localhost binding does.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 4: HTTPS with Caddy
&lt;/h2&gt;

&lt;p&gt;Any reverse proxy works. &lt;a href="https://caddyserver.com/docs/quick-starts/reverse-proxy" rel="noopener noreferrer"&gt;Caddy&lt;/a&gt; is the shortest, because it obtains and renews the certificate for you once DNS points at the server. Install it from your distribution’s packages and put this in &lt;code&gt;/etc/caddy/Caddyfile&lt;/code&gt;:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight go"&gt;&lt;code&gt;&lt;span class="n"&gt;jobs&lt;/span&gt;&lt;span class="o"&gt;.&lt;/span&gt;&lt;span class="n"&gt;example&lt;/span&gt;&lt;span class="o"&gt;.&lt;/span&gt;&lt;span class="n"&gt;com&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;
    &lt;span class="n"&gt;reverse_proxy&lt;/span&gt; &lt;span class="m"&gt;127.0.0.1&lt;/span&gt;&lt;span class="o"&gt;:&lt;/span&gt;&lt;span class="m"&gt;3000&lt;/span&gt;
&lt;span class="p"&gt;}&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Reload Caddy (&lt;code&gt;sudo systemctl reload caddy&lt;/code&gt;) and open &lt;code&gt;https://jobs.example.com&lt;/code&gt;. The Compose file already sets &lt;code&gt;AUTH_TRUST_HOST=true&lt;/code&gt;, so login works behind the proxy; &lt;code&gt;NEXTAUTH_URL&lt;/code&gt; must be the public &lt;code&gt;https://&lt;/code&gt; address. If you use nginx or Traefik instead, proxy to the same &lt;code&gt;127.0.0.1:3000&lt;/code&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 5: Back up db-data and uploads
&lt;/h2&gt;

&lt;p&gt;The two volumes are your whole recruiting history: the database and the CV files. Because Compose names the project &lt;code&gt;vekt&lt;/code&gt;, they appear as &lt;code&gt;vekt_db-data&lt;/code&gt; and &lt;code&gt;vekt_uploads&lt;/code&gt; in &lt;code&gt;docker volume ls&lt;/code&gt;, and the app container is called &lt;code&gt;vekt-app&lt;/code&gt;.&lt;/p&gt;

&lt;p&gt;The database is a single SQLite file, and SQLite warns that copying a database file while a write is in progress can produce a corrupt backup (&lt;a href="https://www.sqlite.org/howtocorrupt.html" rel="noopener noreferrer"&gt;How To Corrupt An SQLite Database File&lt;/a&gt;, section 1.2). The simplest safe method is to stop the app for the few seconds the copy takes. This script follows the pattern in &lt;a href="https://docs.docker.com/engine/storage/volumes/" rel="noopener noreferrer"&gt;Docker’s volume backup guide&lt;/a&gt;, using &lt;code&gt;--volumes-from&lt;/code&gt; to mount both volumes:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;&lt;span class="c"&gt;#!/usr/bin/env bash&lt;/span&gt;
&lt;span class="c"&gt;# /opt/vekt/backup.sh: back up the Vekt database and CV files&lt;/span&gt;
&lt;span class="nb"&gt;set&lt;/span&gt; &lt;span class="nt"&gt;-euo&lt;/span&gt; pipefail
&lt;span class="nb"&gt;cd&lt;/span&gt; /opt/vekt
&lt;span class="nb"&gt;mkdir&lt;/span&gt; &lt;span class="nt"&gt;-p&lt;/span&gt; backups

docker compose stop app
docker run &lt;span class="nt"&gt;--rm&lt;/span&gt; &lt;span class="nt"&gt;--volumes-from&lt;/span&gt; vekt-app &lt;span class="nt"&gt;-v&lt;/span&gt; &lt;span class="s2"&gt;"&lt;/span&gt;&lt;span class="nv"&gt;$PWD&lt;/span&gt;&lt;span class="s2"&gt;/backups:/backup"&lt;/span&gt; alpine &lt;span class="se"&gt;\&lt;/span&gt;
  &lt;span class="nb"&gt;tar &lt;/span&gt;czf &lt;span class="s2"&gt;"/backup/vekt-&lt;/span&gt;&lt;span class="si"&gt;$(&lt;/span&gt;&lt;span class="nb"&gt;date&lt;/span&gt; +%F&lt;span class="si"&gt;)&lt;/span&gt;&lt;span class="s2"&gt;.tar.gz"&lt;/span&gt; /data /app/private/uploads
docker compose start app

&lt;span class="c"&gt;# keep the last 14 archives&lt;/span&gt;
&lt;span class="nb"&gt;ls&lt;/span&gt; &lt;span class="nt"&gt;-1t&lt;/span&gt; backups/vekt-&lt;span class="k"&gt;*&lt;/span&gt;.tar.gz | &lt;span class="nb"&gt;tail&lt;/span&gt; &lt;span class="nt"&gt;-n&lt;/span&gt; +15 | xargs &lt;span class="nt"&gt;-r&lt;/span&gt; &lt;span class="nb"&gt;rm&lt;/span&gt; &lt;span class="nt"&gt;--&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Run it nightly from root’s crontab, away from the 02:00 UTC purge:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;30 3 * * * /opt/vekt/backup.sh &amp;gt;&amp;gt; /var/log/vekt-backup.log 2&amp;gt;&amp;amp;1
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;To restore, stop the app, unpack an archive into the same paths and start it again:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;docker compose stop app
docker run &lt;span class="nt"&gt;--rm&lt;/span&gt; &lt;span class="nt"&gt;--volumes-from&lt;/span&gt; vekt-app &lt;span class="nt"&gt;-v&lt;/span&gt; &lt;span class="s2"&gt;"&lt;/span&gt;&lt;span class="nv"&gt;$PWD&lt;/span&gt;&lt;span class="s2"&gt;/backups:/backup"&lt;/span&gt; alpine &lt;span class="se"&gt;\&lt;/span&gt;
  sh &lt;span class="nt"&gt;-c&lt;/span&gt; &lt;span class="s2"&gt;"cd / &amp;amp;&amp;amp; tar xzf /backup/vekt-2026-11-03.tar.gz"&lt;/span&gt;
docker compose start app
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Two things to keep in mind. The archives contain CVs and personal data, so copy them off the server to storage you control, encrypted, and give them a retention period no longer than you need. And the in-app retention setting only deletes records from the live database: old backups keep deleted candidates until the archives themselves expire, so set the backup rotation to match your retention policy.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 6: Updates
&lt;/h2&gt;

&lt;p&gt;The image is published to &lt;code&gt;ghcr.io/behnoudmst/vekt&lt;/code&gt; on each GitHub release, and &lt;code&gt;latest&lt;/code&gt; points at the newest stable one. To update, take a backup, then pull and restart:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;./backup.sh
docker compose &lt;span class="nt"&gt;-f&lt;/span&gt; docker-compose.yml &lt;span class="nt"&gt;-f&lt;/span&gt; docker-compose.override.yml pull
docker compose &lt;span class="nt"&gt;-f&lt;/span&gt; docker-compose.yml &lt;span class="nt"&gt;-f&lt;/span&gt; docker-compose.override.yml up &lt;span class="nt"&gt;-d&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Database migrations run automatically when the new container starts. If you prefer to choose when you upgrade, set &lt;code&gt;image:&lt;/code&gt; in the override file to a version tag instead of &lt;code&gt;latest&lt;/code&gt;; the release workflow tags each image with its version number, and the &lt;a href="https://github.com/Behnoudmst/vekt/releases" rel="noopener noreferrer"&gt;releases page&lt;/a&gt; lists them.&lt;/p&gt;

&lt;h2&gt;
  
  
  Go-live checklist
&lt;/h2&gt;

&lt;p&gt;Run through this before you publish a real job:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Check&lt;/th&gt;
&lt;th&gt;How&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;HTTPS works and port 3000 is closed&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;https://jobs.example.com&lt;/code&gt; loads; from another machine, &lt;code&gt;curl http://&amp;lt;server-ip&amp;gt;:3000&lt;/code&gt; fails&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Seeding is off&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;SEED_ON_START="false"&lt;/code&gt; and no &lt;code&gt;SEED_ADMIN_PASSWORD&lt;/code&gt; in &lt;code&gt;.env&lt;/code&gt;, then &lt;code&gt;docker compose up -d&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Admin password is your own&lt;/td&gt;
&lt;td&gt;Log in at &lt;code&gt;/login&lt;/code&gt; and confirm the password you chose works&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Recruiters have their own accounts&lt;/td&gt;
&lt;td&gt;Create them in the Admin Dashboard; don’t share the admin login&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Sample job is gone&lt;/td&gt;
&lt;td&gt;Delete “UX/UI Designer” from the recruiter dashboard&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Retention period is set&lt;/td&gt;
&lt;td&gt;Admin Dashboard: data retention (90 days if never set)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Emails are delivered&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;RESEND_API_KEY&lt;/code&gt; and &lt;code&gt;EMAIL_FROM&lt;/code&gt; reach the container (override file), sending domain verified in Resend; apply to a test job with your own address&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Status links point at your domain&lt;/td&gt;
&lt;td&gt;The link in the test email starts with &lt;code&gt;https://jobs.example.com&lt;/code&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;AI provider is the one you mean&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;AI_PROVIDER&lt;/code&gt; set; with &lt;code&gt;mock&lt;/code&gt;, scores are random placeholders&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Privacy notice is yours&lt;/td&gt;
&lt;td&gt;Read &lt;code&gt;/privacy&lt;/code&gt; as a candidate would: controller name, contact address, retention period&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Backups run and restore&lt;/td&gt;
&lt;td&gt;Run &lt;code&gt;backup.sh&lt;/code&gt;, restore into a test install once&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Logs are readable&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;docker compose logs -f app&lt;/code&gt; shows the evaluation of your test application&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;h2&gt;
  
  
  A note on the AI provider and where data goes
&lt;/h2&gt;

&lt;p&gt;With &lt;code&gt;AI_PROVIDER=mock&lt;/code&gt; or &lt;code&gt;ollama&lt;/code&gt;, CV scoring happens on hardware you run, so candidate data stays on your servers. With &lt;code&gt;openai&lt;/code&gt; or &lt;code&gt;openrouter&lt;/code&gt;, the CV text is sent to that provider after Vekt removes names, contact details and other direct identifiers. Candidate emails, whatever the provider, go through Resend.&lt;/p&gt;

&lt;p&gt;The score itself is a reading-order suggestion: it is uncalibrated, and below the threshold candidates are marked &lt;em&gt;needs review&lt;/em&gt; rather than rejected, unless an admin turns on automatic rejection. Self-hosting gives you control over where the data lives; the privacy notice, the lawful basis and answers to candidate requests remain your organisation’s job. More on the platform is on the &lt;a href="https://vekt.eu/" rel="noopener noreferrer"&gt;Vekt home page&lt;/a&gt;, and the code and licence are on our &lt;a href="https://vekt.eu/opensource/" rel="noopener noreferrer"&gt;open-source page&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  FAQ
&lt;/h2&gt;

&lt;h3&gt;
  
  
  What server do I need to self-host an ATS with Docker?
&lt;/h3&gt;

&lt;p&gt;For Vekt, any Linux server that runs Docker with Compose v2. The stack is two containers and a SQLite file, with no separate database server. If you run a local model with Ollama on the same machine, size the server for the model, not for Vekt.&lt;/p&gt;

&lt;h3&gt;
  
  
  Do I need a domain name and HTTPS?
&lt;/h3&gt;

&lt;p&gt;For anything beyond a test, yes. Candidates upload CVs and recruiters log in, so the site should only be reachable over HTTPS. A reverse proxy such as Caddy handles the certificate, and Vekt’s port should stay bound to localhost.&lt;/p&gt;

&lt;h3&gt;
  
  
  Why are no candidate emails sent after setup?
&lt;/h3&gt;

&lt;p&gt;On a stock install, &lt;code&gt;RESEND_API_KEY&lt;/code&gt;, &lt;code&gt;EMAIL_FROM&lt;/code&gt; and &lt;code&gt;APP_URL&lt;/code&gt; are in &lt;code&gt;.env&lt;/code&gt; but not in the Compose file’s list of variables passed to the container. Add them through an override file as shown above, and verify your sending domain in Resend.&lt;/p&gt;

&lt;h3&gt;
  
  
  How do I back up a self-hosted ATS?
&lt;/h3&gt;

&lt;p&gt;Back up both volumes, &lt;code&gt;db-data&lt;/code&gt; (the SQLite database) and &lt;code&gt;uploads&lt;/code&gt; (the CV files), with the app stopped for the few seconds the copy takes, and keep the archives encrypted and off the server. Test a restore at least once.&lt;/p&gt;

&lt;h3&gt;
  
  
  Is Vekt free to self-host?
&lt;/h3&gt;

&lt;p&gt;Yes. Vekt is MIT-licensed, with no seat or job limits. Your costs are the server, your time to run it, and any AI provider or email service you choose to use.&lt;/p&gt;

</description>
      <category>selfhosted</category>
      <category>docker</category>
      <category>tutorial</category>
    </item>
    <item>
      <title>Self-hosted Greenhouse alternative: when to run your own ATS</title>
      <dc:creator>Behnoud mostafaie</dc:creator>
      <pubDate>Fri, 30 Oct 2026 00:00:00 +0000</pubDate>
      <link>https://dev.to/vekt/self-hosted-greenhouse-alternative-when-to-run-your-own-ats-5i1</link>
      <guid>https://dev.to/vekt/self-hosted-greenhouse-alternative-when-to-run-your-own-ats-5i1</guid>
      <description>&lt;p&gt;If you are looking for a &lt;strong&gt;self-hosted Greenhouse alternative&lt;/strong&gt; , the first thing to know is that Greenhouse is a cloud product only: there is no version you can run on your own servers. So the real question is not which tool copies Greenhouse, but whether running your own ATS fits your team better than renting a very complete one. This guide answers that honestly, from the people who build &lt;a href="https://vekt.eu/self-hosted-ats/" rel="noopener noreferrer"&gt;Vekt&lt;/a&gt;, an open-source recruiting platform you host yourself.&lt;/p&gt;

&lt;p&gt;We read Greenhouse’s own &lt;a href="https://www.greenhouse.com/pricing" rel="noopener noreferrer"&gt;pricing page&lt;/a&gt;, &lt;a href="https://www.greenhouse.com/ats" rel="noopener noreferrer"&gt;ATS page&lt;/a&gt;, &lt;a href="https://www.greenhouse.com/ai-recruiting" rel="noopener noreferrer"&gt;AI recruiting page&lt;/a&gt;, &lt;a href="https://www.greenhouse.com/interviewing-decision-making" rel="noopener noreferrer"&gt;interviewing page&lt;/a&gt;, &lt;a href="https://support.greenhouse.io/hc/en-us/articles/4416552708507-Greenhouse-security-overview" rel="noopener noreferrer"&gt;security overview&lt;/a&gt; and &lt;a href="https://integrations.greenhouse.com/" rel="noopener noreferrer"&gt;partner directory&lt;/a&gt; on 30 October 2026. Every Greenhouse fact below comes from one of those pages, and every Vekt fact from &lt;a href="https://github.com/Behnoudmst/vekt" rel="noopener noreferrer"&gt;its repository&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  The short answer
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Stay on Greenhouse&lt;/strong&gt; if you hire at volume, run structured interview loops across many teams, and rely on scheduling, sourcing, reporting and hundreds of integrations. Vekt does not replace that today.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Consider self-hosting&lt;/strong&gt; if you hire a handful of roles at a time, want candidate data in infrastructure you control, want to choose (or run) the AI model that reads CVs, and want code you can read and change under the MIT licence.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  What Greenhouse is
&lt;/h2&gt;

&lt;p&gt;Greenhouse is one of the best-known applicant tracking systems. It sells three plans, Core, Plus and Pro, and does not publish prices: its pricing page says “Greenhouse pricing is customized based on your company’s hiring needs”, and you get a quote after a demo.&lt;/p&gt;

&lt;p&gt;Even the entry plan is broad. According to the &lt;a href="https://www.greenhouse.com/pricing" rel="noopener noreferrer"&gt;pricing page&lt;/a&gt;, Core includes sourcing and CRM, structured interview kits and scorecards, an AI notetaker, scheduling, reporting and analytics, single sign-on and talent matching. Plus adds texting, a business intelligence connector, sourcing automation and fraud and spam detection. Pro adds an audit log, a developer sandbox and enterprise data configuration.&lt;/p&gt;

&lt;p&gt;Greenhouse also takes care with AI. Its &lt;a href="https://www.greenhouse.com/ai-recruiting" rel="noopener noreferrer"&gt;AI recruiting page&lt;/a&gt; says “the decision is always yours, not handed to an algorithm”, says customers can switch any AI feature on or off for the whole organisation, and describes monthly third-party bias audits and resume review with identifying details redacted. It states that it does not train its AI on personal data.&lt;/p&gt;

&lt;h2&gt;
  
  
  What Greenhouse does that Vekt doesn’t
&lt;/h2&gt;

&lt;p&gt;Greenhouse is a much larger product, and you should know exactly where before you consider moving. Vekt has none of the following today:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Interview scheduling.&lt;/strong&gt; Greenhouse offers candidate self-scheduling, scheduling links, interviewer limits and AI scheduling suggestions, and works with Zoom, Google Meet, Microsoft Teams and Spark Hire (&lt;a href="https://www.greenhouse.com/ats" rel="noopener noreferrer"&gt;ATS page&lt;/a&gt;, &lt;a href="https://www.greenhouse.com/interviewing-decision-making" rel="noopener noreferrer"&gt;interviewing page&lt;/a&gt;).&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Structured interview kits and scorecards.&lt;/strong&gt; A core part of Greenhouse’s approach to structured hiring. Vekt structures the application (screening questions, an AI-ranked reading order) but not the interviews that follow.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Interview notes and AI voice interviews.&lt;/strong&gt; Greenhouse records and transcribes interviews and maps the notes to scorecard questions, and offers an AI voice first-round interview.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Sourcing and CRM.&lt;/strong&gt; Greenhouse manages prospects who haven’t applied yet and recommends candidates when a role opens.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Reporting.&lt;/strong&gt; Greenhouse lists “40+ pre-built reports and dashboards”, custom reports and BI connectors. Vekt shows applications per job sorted by score and records job page views, but has no reporting module.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Integrations.&lt;/strong&gt; Greenhouse’s &lt;a href="https://integrations.greenhouse.com/" rel="noopener noreferrer"&gt;partner directory&lt;/a&gt; listed 520 entries when we checked, across HRIS and onboarding, job distribution, productivity and candidate experience. Vekt has a documented REST API (Swagger UI at &lt;code&gt;/api-docs&lt;/code&gt;) and no ready-made integrations.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Job-board distribution, texting and onboarding.&lt;/strong&gt; Greenhouse has all three; Vekt publishes jobs on its own careers page only and sends candidate emails.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Single sign-on and enterprise controls.&lt;/strong&gt; SSO is in Greenhouse’s Core plan; an audit log and sandbox are in Pro. Vekt signs users in with email and password and has two roles, admin and recruiter.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Certifications and a vendor behind it.&lt;/strong&gt; Greenhouse’s ATS page cites ISO/IEC 42001 and SOC and ISO certifications, and you have a company to call. Vekt is a young open-source project with no certifications and no support contract.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  What changes when you self-host instead
&lt;/h2&gt;

&lt;p&gt;Self-hosting moves both control and work to you. Here is what that means with Vekt specifically.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Where the data lives.&lt;/strong&gt; Greenhouse runs on AWS, with US, EU (Frankfurt, with Ireland for disaster recovery) and APAC silos according to its &lt;a href="https://support.greenhouse.io/hc/en-us/articles/4416552708507-Greenhouse-security-overview" rel="noopener noreferrer"&gt;security overview&lt;/a&gt;. With Vekt, the database (SQLite) and uploaded CVs sit in Docker volumes on a server you choose. Candidate data stays on your servers when scoring runs on a local model through Ollama; if you pick OpenAI or OpenRouter, the redacted CV text goes to that provider, and candidate emails are sent through Resend.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Which AI reads the CV.&lt;/strong&gt; You set &lt;code&gt;AI_PROVIDER&lt;/code&gt; to &lt;code&gt;openai&lt;/code&gt;, &lt;code&gt;openrouter&lt;/code&gt;, &lt;code&gt;ollama&lt;/code&gt; or &lt;code&gt;mock&lt;/code&gt;. Before any model sees a CV, Vekt removes the name, email, phone, links, date of birth and declared personal attributes. Every evaluation stores the provider, model, prompt hash and evaluator version.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;How AI affects decisions.&lt;/strong&gt; A candidate scoring below the threshold is marked &lt;em&gt;needs review&lt;/em&gt;, not rejected, and gets no email until a recruiter decides. Automatic rejection exists only as an admin opt-in (&lt;code&gt;AUTO_REJECT_BELOW_THRESHOLD&lt;/code&gt;). Status emails wait 48 hours by default and are cancelled if a recruiter changes the decision. The scores themselves are uncalibrated and have never been measured against hiring outcomes; we explain that in &lt;a href="https://vekt.eu/blog/what-vekts-ai-cant-do/" rel="noopener noreferrer"&gt;what Vekt’s AI can’t do&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Retention.&lt;/strong&gt; A daily job deletes candidates and their CV files once the retention period set by the admin has passed. You decide the number; Vekt enforces it.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Cost.&lt;/strong&gt; The software is free under the MIT licence with no per-seat pricing. You pay for a server, your time to run it, and the AI provider’s usage if you don’t use a local model. With Greenhouse you pay a quoted subscription and the vendor runs everything.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Your new jobs.&lt;/strong&gt; Updates, backups of the &lt;code&gt;db-data&lt;/code&gt; and &lt;code&gt;uploads&lt;/code&gt; volumes, HTTPS, and keeping secrets in &lt;code&gt;.env&lt;/code&gt; are on you. Self-hosting doesn’t make your hiring lawful on its own either: your privacy notice, lawful basis and answers to candidate requests remain your organisation’s work.&lt;/p&gt;

&lt;h2&gt;
  
  
  Feature comparison
&lt;/h2&gt;

&lt;p&gt;Built only from the pages listed above (Greenhouse) and the Vekt repository, checked 30 October 2026.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;&lt;/th&gt;
&lt;th&gt;Greenhouse&lt;/th&gt;
&lt;th&gt;Vekt&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Hosting&lt;/td&gt;
&lt;td&gt;Vendor cloud on AWS (US, EU and APAC silos)&lt;/td&gt;
&lt;td&gt;Your server, Docker Compose&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Licence and source&lt;/td&gt;
&lt;td&gt;Proprietary&lt;/td&gt;
&lt;td&gt;MIT, source on GitHub&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Pricing&lt;/td&gt;
&lt;td&gt;Quote only, three plans (Core, Plus, Pro)&lt;/td&gt;
&lt;td&gt;Free software; you pay hosting and any AI provider&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Job posting&lt;/td&gt;
&lt;td&gt;Multilingual job boards; job-distribution partners in its directory&lt;/td&gt;
&lt;td&gt;Own careers page with readable job URLs&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Screening questions on apply&lt;/td&gt;
&lt;td&gt;Not checked on the pages we opened&lt;/td&gt;
&lt;td&gt;Single- and multiple-choice, per job&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;AI CV review&lt;/td&gt;
&lt;td&gt;Yes, with identifying details redacted&lt;/td&gt;
&lt;td&gt;Yes, with direct identifiers removed first&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Choice of AI model&lt;/td&gt;
&lt;td&gt;Not offered on the pages we opened; AI features can be turned off&lt;/td&gt;
&lt;td&gt;OpenAI, OpenRouter or a local model via Ollama&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Low AI score&lt;/td&gt;
&lt;td&gt;“The decision is always yours”&lt;/td&gt;
&lt;td&gt;Marked &lt;em&gt;needs review&lt;/em&gt;; auto-reject only as an opt-in&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Interview scheduling&lt;/td&gt;
&lt;td&gt;Yes, including self-scheduling&lt;/td&gt;
&lt;td&gt;No&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Interview kits and scorecards&lt;/td&gt;
&lt;td&gt;Yes&lt;/td&gt;
&lt;td&gt;No&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Sourcing and CRM&lt;/td&gt;
&lt;td&gt;Yes&lt;/td&gt;
&lt;td&gt;No&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Reporting&lt;/td&gt;
&lt;td&gt;40+ pre-built reports, BI connector (Plus)&lt;/td&gt;
&lt;td&gt;Applications per job, job page views&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Integrations&lt;/td&gt;
&lt;td&gt;520 entries in the partner directory&lt;/td&gt;
&lt;td&gt;REST API, no ready-made integrations&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;SSO&lt;/td&gt;
&lt;td&gt;Yes (Core)&lt;/td&gt;
&lt;td&gt;No&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Audit log&lt;/td&gt;
&lt;td&gt;Pro plan&lt;/td&gt;
&lt;td&gt;Evaluation provenance per score; no general audit log&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Automatic data deletion&lt;/td&gt;
&lt;td&gt;Not checked on the pages we opened&lt;/td&gt;
&lt;td&gt;Daily purge after the retention period you set&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Onboarding&lt;/td&gt;
&lt;td&gt;Yes&lt;/td&gt;
&lt;td&gt;No&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;“Not checked on the pages we opened” means exactly that: the feature may well exist in Greenhouse, but we only list what we could link to.&lt;/p&gt;

&lt;h2&gt;
  
  
  Who should stay on Greenhouse
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Teams with recruiting coordinators who live in interview scheduling and scorecards.&lt;/li&gt;
&lt;li&gt;Companies hiring dozens of roles at once, with sourcing, CRM and reporting in daily use.&lt;/li&gt;
&lt;li&gt;Organisations whose HRIS, job boards and assessments already connect through Greenhouse’s integrations.&lt;/li&gt;
&lt;li&gt;Anyone who needs SSO, certifications and a vendor contract on day one, and has no one to run a server.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;For those teams, moving to any self-hosted tool would be a step backwards today.&lt;/p&gt;

&lt;h2&gt;
  
  
  Who self-hosting suits
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Small and mid-sized teams hiring a few roles at a time, for whom a full Greenhouse plan is more than they use.&lt;/li&gt;
&lt;li&gt;EU teams that want candidate data on infrastructure they choose, with a local model so CVs never leave it.&lt;/li&gt;
&lt;li&gt;Teams that want AI to set a reading order while a person makes every rejection.&lt;/li&gt;
&lt;li&gt;Developers who want to change the hiring logic itself. Vekt is a TypeScript codebase; see &lt;a href="https://vekt.eu/opensource/" rel="noopener noreferrer"&gt;our open-source page&lt;/a&gt;.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;If you want to compare Vekt with other open-source options rather than with Greenhouse, our &lt;a href="https://vekt.eu/blog/open-source-ats-compared/" rel="noopener noreferrer"&gt;open-source ATS comparison&lt;/a&gt; covers nine of them, including where another project fits better.&lt;/p&gt;

&lt;h2&gt;
  
  
  Try it before you decide
&lt;/h2&gt;

&lt;p&gt;The Vekt README offers a one-line setup that downloads the files, generates secrets, asks for admin credentials and starts the app and Inngest with Docker:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;curl &lt;span class="nt"&gt;-fsSL&lt;/span&gt; https://raw.githubusercontent.com/Behnoudmst/vekt/main/setup.sh | bash
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;It needs Docker with Compose v2 and &lt;code&gt;openssl&lt;/code&gt;. The default &lt;code&gt;AI_PROVIDER&lt;/code&gt; is &lt;code&gt;mock&lt;/code&gt;, so nothing leaves the machine until you choose a provider. Post one real job, run it next to your current process for a week, and check what you miss. More on what the platform covers is on the &lt;a href="https://vekt.eu/" rel="noopener noreferrer"&gt;Vekt home page&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  FAQ
&lt;/h2&gt;

&lt;h3&gt;
  
  
  Can you self-host Greenhouse?
&lt;/h3&gt;

&lt;p&gt;No. Greenhouse is sold as a cloud service only. Its security overview describes hosting on AWS in US, EU and APAC regions run by Greenhouse. If you need to run the ATS on your own servers, you need a different product.&lt;/p&gt;

&lt;h3&gt;
  
  
  Is there an open-source alternative to Greenhouse?
&lt;/h3&gt;

&lt;p&gt;Several open-source ATS projects exist, including OpenCATS, FreeATS and Vekt, and HR suites such as Horilla and OrangeHRM include recruitment. None matches Greenhouse feature for feature; our &lt;a href="https://vekt.eu/blog/open-source-ats-compared/" rel="noopener noreferrer"&gt;comparison&lt;/a&gt; shows what each covers.&lt;/p&gt;

&lt;h3&gt;
  
  
  How much does Greenhouse cost?
&lt;/h3&gt;

&lt;p&gt;Greenhouse doesn’t publish prices. Its pricing page says pricing “is customized based on your company’s hiring needs” across the Core, Plus and Pro plans, and a quote comes after a demo.&lt;/p&gt;

&lt;h3&gt;
  
  
  Does Vekt do interview scheduling like Greenhouse?
&lt;/h3&gt;

&lt;p&gt;Not today. Vekt covers jobs, the careers page, applications with screening questions, AI-ranked review and candidate emails. Scheduling, interview kits and scorecards are not part of it yet.&lt;/p&gt;

&lt;h3&gt;
  
  
  Does self-hosting keep CVs away from AI providers?
&lt;/h3&gt;

&lt;p&gt;Only if you run the model yourself. With &lt;code&gt;AI_PROVIDER=ollama&lt;/code&gt;, scoring happens on your hardware. With OpenAI or OpenRouter, the CV text, with direct identifiers removed, is sent to that provider.&lt;/p&gt;

</description>
      <category>greenhousealternativ</category>
      <category>selfhosted</category>
      <category>comparison</category>
    </item>
    <item>
      <title>Why we publish what Vekt's AI can't do</title>
      <dc:creator>Behnoud mostafaie</dc:creator>
      <pubDate>Tue, 27 Oct 2026 09:54:38 +0000</pubDate>
      <link>https://dev.to/behnoudmst/why-we-publish-what-vekts-ai-cant-do-4kln</link>
      <guid>https://dev.to/behnoudmst/why-we-publish-what-vekts-ai-cant-do-4kln</guid>
      <description>&lt;p&gt;Vekt is an open-source recruitment orchestration engine for privacy-first hiring. It runs on your own servers, structures how candidates are evaluated, and uses AI to rank applications against the job description so a recruiter knows what to read first. It does not decide who gets hired.&lt;/p&gt;

&lt;p&gt;Before the feature list, the repo has a &lt;a href="https://github.com/Behnoudmst/vekt/tree/main/docs/compliance" rel="noopener noreferrer"&gt;compliance folder&lt;/a&gt; that names what the AI scoring can't do. This post explains those four limits and why they come first.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why lead with limits
&lt;/h2&gt;

&lt;p&gt;In the EU, using AI to screen job applications is classed as a high-risk use under Annex III of the AI Act. Those obligations now apply from 2 December 2027, after the Digital Omnibus moved the original date. GDPR already applies today: candidates have rights over decisions based solely on automated processing, and they have to be told how their data is used.&lt;/p&gt;

&lt;p&gt;It is easy to answer that with a "compliant" badge. We think a tool in this space should be easier to question instead, and that starts with saying plainly where it falls short.&lt;/p&gt;

&lt;h2&gt;
  
  
  Limit 1: no accuracy has been measured
&lt;/h2&gt;

&lt;p&gt;Vekt has never been tested against a labelled set of CVs and hiring outcomes. There is no precision, recall or agreement-with-reviewer figure. Any number we gave you today would be invented.&lt;/p&gt;

&lt;p&gt;If a vendor gives you an accuracy figure, ask what it was measured on, and whether it was measured on hiring like yours.&lt;/p&gt;

&lt;h2&gt;
  
  
  Limit 2: redaction removes names, not every clue
&lt;/h2&gt;

&lt;p&gt;Before any text reaches the model, Vekt removes direct identifiers: name, email, phone, links, date of birth, and lines that state nationality, citizenship, marital status or gender.&lt;/p&gt;

&lt;p&gt;It does not remove proxies. University names, languages spoken, military service, career gaps and writing style can all carry signals of origin, age and gender, and a language model can pick them up. Redaction reduces the risk; it does not remove it. That is why a person makes the final call.&lt;/p&gt;

&lt;h2&gt;
  
  
  Limit 3: scores aren't comparable across setups
&lt;/h2&gt;

&lt;p&gt;The score comes from a general-purpose language model asked for a number from 0 to 100. Change the model, the provider or the prompt, and the same CV gets a different number. Even the same CV scored twice can drift by a few points.&lt;/p&gt;

&lt;p&gt;So scores are only comparable within one configuration, and the threshold you choose sits on a noisy signal. Write down why you picked it.&lt;/p&gt;

&lt;h2&gt;
  
  
  Limit 4: running it yourself doesn't make you compliant
&lt;/h2&gt;

&lt;p&gt;Self-hosting answers one important question: where the data is. With a local model, CV text never leaves infrastructure you control. With a hosted provider such as OpenAI or OpenRouter, the redacted text does go to that provider, which becomes a processor and possibly a transfer outside the EU.&lt;/p&gt;

&lt;p&gt;What self-hosting does not answer is whether your hiring process is lawful and fair. That depends on your criteria, your human review, what you tell candidates and how you handle their requests. Those stay with the organisation that hires.&lt;/p&gt;

&lt;h2&gt;
  
  
  What we're doing about it
&lt;/h2&gt;

&lt;p&gt;The biggest gap is the first one. We're looking for two or three EU organisations willing to measure Vekt on their own past applications, on their own servers, with their data never leaving them. We'll help set it up, and with your agreement the method and aggregate results get published, so the next team doesn't start from zero.&lt;/p&gt;

&lt;p&gt;If that's you, &lt;a href="https://github.com/Behnoudmst/vekt/issues" rel="noopener noreferrer"&gt;open an issue on GitHub&lt;/a&gt; or message us on LinkedIn.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Originally published on &lt;a href="https://vekt.eu/blog/what-vekts-ai-cant-do/" rel="noopener noreferrer"&gt;vekt.eu&lt;/a&gt;. Engineering view, not legal advice. What Vekt does and does not do is documented in the &lt;a href="https://github.com/Behnoudmst/vekt/tree/main/docs/compliance" rel="noopener noreferrer"&gt;compliance folder&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>ai</category>
      <category>opensource</category>
      <category>privacy</category>
      <category>hiring</category>
    </item>
    <item>
      <title>Open-source and self-hosted ATS options in 2026, compared</title>
      <dc:creator>Behnoud mostafaie</dc:creator>
      <pubDate>Sun, 25 Oct 2026 00:00:00 +0000</pubDate>
      <link>https://dev.to/vekt/open-source-and-self-hosted-ats-options-in-2026-compared-5g85</link>
      <guid>https://dev.to/vekt/open-source-and-self-hosted-ats-options-in-2026-compared-5g85</guid>
      <description>&lt;p&gt;If you want an &lt;strong&gt;open-source ATS&lt;/strong&gt; you can &lt;strong&gt;self-host&lt;/strong&gt; , you have more choice in 2026 than most roundups suggest, and the options differ more than their feature lists show. Some are dedicated applicant tracking systems, some are recruitment modules inside a full HR or ERP suite, and a few are new AI-first projects. Their licences range from MIT to AGPL, which matters if you plan to change the code.&lt;/p&gt;

&lt;p&gt;This guide compares nine of them. We wrote it as the people behind &lt;a href="https://vekt.eu/self-hosted-ats/" rel="noopener noreferrer"&gt;Vekt&lt;/a&gt;, one of the nine, so read our section with that in mind. To keep it fair, every licence below was read from the project’s own repository on 25 October 2026, every other fact links to its source, and each section says who the project suits best, including where that isn’t us.&lt;/p&gt;

&lt;h2&gt;
  
  
  The short answer
&lt;/h2&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Project&lt;/th&gt;
&lt;th&gt;Best for&lt;/th&gt;
&lt;th&gt;Licence&lt;/th&gt;
&lt;th&gt;Stack&lt;/th&gt;
&lt;th&gt;Scope&lt;/th&gt;
&lt;th&gt;Last commit on main*&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;a href="https://github.com/opencats/OpenCATS" rel="noopener noreferrer"&gt;OpenCATS&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;A classic ATS on PHP hosting&lt;/td&gt;
&lt;td&gt;MPL 2.0&lt;/td&gt;
&lt;td&gt;PHP, MySQL&lt;/td&gt;
&lt;td&gt;Dedicated ATS&lt;/td&gt;
&lt;td&gt;22 Oct 2026&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;a href="https://github.com/freeats/freeats" rel="noopener noreferrer"&gt;FreeATS&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;A conventional ATS with an optional hosted plan&lt;/td&gt;
&lt;td&gt;MIT&lt;/td&gt;
&lt;td&gt;Ruby on Rails, Docker&lt;/td&gt;
&lt;td&gt;Dedicated ATS&lt;/td&gt;
&lt;td&gt;13 Oct 2026&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;a href="https://github.com/Behnoudmst/vekt" rel="noopener noreferrer"&gt;Vekt&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;AI screening with human review and data deletion, EU teams&lt;/td&gt;
&lt;td&gt;MIT&lt;/td&gt;
&lt;td&gt;Next.js, SQLite, Docker&lt;/td&gt;
&lt;td&gt;Recruiting platform&lt;/td&gt;
&lt;td&gt;18 Oct 2026&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;a href="https://github.com/maxig/vellum-oss" rel="noopener noreferrer"&gt;Vellum&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;An AI-first ATS with a shared inbox&lt;/td&gt;
&lt;td&gt;AGPL 3.0&lt;/td&gt;
&lt;td&gt;Next.js, PostgreSQL, Docker&lt;/td&gt;
&lt;td&gt;Dedicated ATS&lt;/td&gt;
&lt;td&gt;4 Aug 2026&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;a href="https://github.com/horilla-opensource/horilla" rel="noopener noreferrer"&gt;Horilla&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;One open-source system for all of HR&lt;/td&gt;
&lt;td&gt;LGPL 2.1&lt;/td&gt;
&lt;td&gt;Django, PostgreSQL, Docker&lt;/td&gt;
&lt;td&gt;HR suite&lt;/td&gt;
&lt;td&gt;21 Oct 2026&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;a href="https://github.com/frappe/hrms" rel="noopener noreferrer"&gt;Frappe HR&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;Teams already on ERPNext&lt;/td&gt;
&lt;td&gt;GPL 3.0&lt;/td&gt;
&lt;td&gt;Frappe (Python)&lt;/td&gt;
&lt;td&gt;HR suite&lt;/td&gt;
&lt;td&gt;25 Oct 2026&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;a href="https://github.com/orangehrm/orangehrm" rel="noopener noreferrer"&gt;OrangeHRM&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;An established HRIS with recruitment&lt;/td&gt;
&lt;td&gt;GPL 3.0&lt;/td&gt;
&lt;td&gt;PHP&lt;/td&gt;
&lt;td&gt;HR suite&lt;/td&gt;
&lt;td&gt;29 Jun 2026&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;a href="https://github.com/odoo/odoo" rel="noopener noreferrer"&gt;Odoo Recruitment&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;Teams already on Odoo&lt;/td&gt;
&lt;td&gt;LGPL 3.0 (Community)&lt;/td&gt;
&lt;td&gt;Python, PostgreSQL&lt;/td&gt;
&lt;td&gt;ERP module&lt;/td&gt;
&lt;td&gt;Very active&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;a href="https://github.com/OSSAdmiral/Recruit" rel="noopener noreferrer"&gt;Recruit (OSSAdmiral)&lt;/a&gt;&lt;/td&gt;
&lt;td&gt;Laravel developers who want a base to build on&lt;/td&gt;
&lt;td&gt;AGPL 3.0&lt;/td&gt;
&lt;td&gt;Laravel, Filament&lt;/td&gt;
&lt;td&gt;Dedicated ATS&lt;/td&gt;
&lt;td&gt;13 Nov 2024&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;*Date of the latest commit on the default branch, checked 25 October 2026. Odoo’s main repository was not checked commit by commit; it receives changes daily.&lt;/p&gt;

&lt;h2&gt;
  
  
  How we compared them
&lt;/h2&gt;

&lt;p&gt;Five questions decide whether a self-hosted ATS works for a team:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Licence.&lt;/strong&gt; Can you change the code without publishing your changes? MIT, MPL and LGPL are permissive enough for most in-house changes; GPL and AGPL ask you to share modified source in some situations (see the licence note below).&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Scope.&lt;/strong&gt; Do you want a dedicated ATS, or recruitment as one module of an HR or ERP system?&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Hosting effort.&lt;/strong&gt; Is there a maintained Docker setup, or do you assemble a LAMP or Python stack yourself?&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;AI and data.&lt;/strong&gt; If the tool uses AI on CVs, which providers can it use, can it run on a local model, and what happens to the data?&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Upkeep.&lt;/strong&gt; Is the project actively maintained, and does it say whether it is ready for production?&lt;/li&gt;
&lt;/ol&gt;

&lt;h2&gt;
  
  
  OpenCATS
&lt;/h2&gt;

&lt;p&gt;&lt;a href="https://www.opencats.org/" rel="noopener noreferrer"&gt;OpenCATS&lt;/a&gt; is the name most people know when they search for an open-source ATS. It is a dedicated applicant tracking system written in PHP with MySQL, and it includes a careers page. The repository has a &lt;code&gt;docker&lt;/code&gt; folder, and the project is still receiving commits in October 2026.&lt;/p&gt;

&lt;p&gt;The licence is split: OpenCATS code is under the Mozilla Public License 2.0, while the original CATS code it grew from is under its own licence (both explained in the repository’s &lt;code&gt;LICENSE.md&lt;/code&gt;).&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Best for:&lt;/strong&gt; teams and agencies that want a classic ATS feature set and are comfortable hosting PHP. Try a test install first to check that its interface fits how your recruiters work today.&lt;/p&gt;

&lt;h2&gt;
  
  
  FreeATS
&lt;/h2&gt;

&lt;p&gt;&lt;a href="https://www.freeats.com/" rel="noopener noreferrer"&gt;FreeATS&lt;/a&gt; is an MIT-licensed ATS built with Ruby on Rails, with a Docker Compose setup for self-hosting. It also sells a hosted version: on its site today, the free cloud plan allows 3 open positions and 3,000 candidate profiles with unlimited users, and Pro costs $99 a month.&lt;/p&gt;

&lt;p&gt;Its feature list is that of a conventional ATS: custom pipelines, scorecards, a LinkedIn import browser extension, email sync, tasks, duplicate detection and a career site.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Best for:&lt;/strong&gt; small teams that want a familiar ATS and like having a hosted plan to fall back on.&lt;/p&gt;

&lt;h2&gt;
  
  
  Vekt
&lt;/h2&gt;

&lt;p&gt;&lt;a href="https://vekt.eu/self-hosted-ats/" rel="noopener noreferrer"&gt;Vekt&lt;/a&gt; is our project: an MIT-licensed, &lt;a href="https://vekt.eu/" rel="noopener noreferrer"&gt;self-hosted recruiting platform&lt;/a&gt; built with Next.js, SQLite and &lt;a href="https://www.inngest.com/" rel="noopener noreferrer"&gt;Inngest&lt;/a&gt; for background jobs, deployed with Docker Compose. It covers job listings, a careers page, applications with screening questions, a candidate pipeline and candidate emails.&lt;/p&gt;

&lt;p&gt;What sets it apart is how it handles AI and personal data. CVs can be scored by OpenAI, OpenRouter or a local model through Ollama. Names, contact details and other direct identifiers are removed from the CV text before any model sees it. A low score never rejects anyone on its own: those candidates are marked &lt;em&gt;needs review&lt;/em&gt; until a recruiter decides. Every evaluation records the provider, model, prompt hash and evaluator version, and a daily job deletes candidates and their CVs once the retention period you set has passed.&lt;/p&gt;

&lt;p&gt;We also publish &lt;a href="https://vekt.eu/blog/what-vekts-ai-cant-do/" rel="noopener noreferrer"&gt;what its AI can’t do&lt;/a&gt;: the scores have never been measured against real hiring outcomes, so Vekt claims no accuracy figure.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Best for:&lt;/strong&gt; teams hiring in the EU that want AI screening with a human in charge, candidate data on their own servers, and code they can read.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Not a fit:&lt;/strong&gt; if you need interview scheduling, job-board multiposting or single sign-on today. Vekt has none of these yet.&lt;/p&gt;

&lt;h2&gt;
  
  
  Vellum
&lt;/h2&gt;

&lt;p&gt;&lt;a href="https://github.com/maxig/vellum-oss" rel="noopener noreferrer"&gt;Vellum&lt;/a&gt; describes itself as “an open-source, AI-first applicant tracking system”. It is built with Next.js, PostgreSQL and Prisma, and ships a Docker Compose setup that includes a reverse proxy. Its README lists a hiring pipeline, branded career sites, a shared inbox, multi-tenancy, and “bring-your-own (or local) AI”.&lt;/p&gt;

&lt;p&gt;The README marks its own status as an “actively developed MVP”. It is licensed under the AGPL 3.0.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Best for:&lt;/strong&gt; teams that want a modern, AI-first ATS with a shared inbox and are comfortable with an MVP and the AGPL’s terms.&lt;/p&gt;

&lt;h2&gt;
  
  
  Horilla
&lt;/h2&gt;

&lt;p&gt;&lt;a href="https://www.horilla.com/features/recruitment/" rel="noopener noreferrer"&gt;Horilla&lt;/a&gt; is a complete open-source HR system: recruitment sits next to onboarding, employee records, attendance, leave, payroll, helpdesk and more. It is built with Django and PostgreSQL, has Docker Compose files, and is licensed under the LGPL 2.1.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Best for:&lt;/strong&gt; organisations that want one open-source system for all of HR, with recruitment as one part of it.&lt;/p&gt;

&lt;h2&gt;
  
  
  Frappe HR
&lt;/h2&gt;

&lt;p&gt;&lt;a href="https://github.com/frappe/hrms" rel="noopener noreferrer"&gt;Frappe HR&lt;/a&gt; is the HR and payroll app of the Frappe and ERPNext ecosystem, licensed under the GPL 3.0. Recruitment covers job openings, applicants and interviews, inside the same system as the rest of HR.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Best for:&lt;/strong&gt; companies already running ERPNext, or happy to adopt the Frappe framework.&lt;/p&gt;

&lt;h2&gt;
  
  
  OrangeHRM
&lt;/h2&gt;

&lt;p&gt;&lt;a href="https://github.com/orangehrm/orangehrm" rel="noopener noreferrer"&gt;OrangeHRM&lt;/a&gt; is a long-established HR system written in PHP. The open-source edition, under the GPL 3.0, includes a recruitment module; the company also sells paid editions with more features.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Best for:&lt;/strong&gt; teams that need an HR information system first and recruitment second.&lt;/p&gt;

&lt;h2&gt;
  
  
  Odoo Recruitment
&lt;/h2&gt;

&lt;p&gt;Recruitment is one of the apps in &lt;a href="https://github.com/odoo/odoo" rel="noopener noreferrer"&gt;Odoo&lt;/a&gt;, the open-source ERP. The Community edition is licensed under the LGPL 3.0; some features are part of the paid Enterprise edition only.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Best for:&lt;/strong&gt; businesses already running Odoo, where recruiting next to accounting, sales and HR outweighs having a dedicated ATS.&lt;/p&gt;

&lt;h2&gt;
  
  
  Recruit (OSSAdmiral)
&lt;/h2&gt;

&lt;p&gt;&lt;a href="https://github.com/OSSAdmiral/Recruit" rel="noopener noreferrer"&gt;Recruit&lt;/a&gt; is a corporate recruitment system built with Laravel and FilamentPHP, under the AGPL 3.0. Its README says it is “under hard development and not yet ready for production”, and its last commit on the default branch was in November 2024.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Best for:&lt;/strong&gt; Laravel developers who want a starting point to build their own recruiting tool, rather than a finished product.&lt;/p&gt;

&lt;h2&gt;
  
  
  What the licences mean in practice
&lt;/h2&gt;

&lt;p&gt;All nine are open source, but they ask different things of you if you change the code:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Licence&lt;/th&gt;
&lt;th&gt;Who uses it here&lt;/th&gt;
&lt;th&gt;If you modify the code and run it internally&lt;/th&gt;
&lt;th&gt;If you offer your modified version to others over the web&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;MIT&lt;/td&gt;
&lt;td&gt;Vekt, FreeATS&lt;/td&gt;
&lt;td&gt;No obligations beyond keeping the notice&lt;/td&gt;
&lt;td&gt;Same: no obligation to publish&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;MPL 2.0&lt;/td&gt;
&lt;td&gt;OpenCATS&lt;/td&gt;
&lt;td&gt;Fine&lt;/td&gt;
&lt;td&gt;Changed files must stay under the MPL if you distribute them&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;LGPL 2.1 / 3.0&lt;/td&gt;
&lt;td&gt;Horilla, Odoo Community&lt;/td&gt;
&lt;td&gt;Fine&lt;/td&gt;
&lt;td&gt;Mainly affects distributing modified copies&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;GPL 3.0&lt;/td&gt;
&lt;td&gt;Frappe HR, OrangeHRM&lt;/td&gt;
&lt;td&gt;Fine&lt;/td&gt;
&lt;td&gt;Distributed copies must include the source&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;AGPL 3.0&lt;/td&gt;
&lt;td&gt;Vellum, Recruit&lt;/td&gt;
&lt;td&gt;Fine&lt;/td&gt;
&lt;td&gt;You must offer the modified source to the users of your hosted service&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;This is a summary, not legal advice. If you plan to build a product on top of any of these, read the licence itself.&lt;/p&gt;

&lt;h2&gt;
  
  
  A note on “GDPR-compliant” labels
&lt;/h2&gt;

&lt;p&gt;Some ATS products, open source and hosted, describe themselves as GDPR-compliant. No software is compliant on its own. Self-hosting lets you decide where candidate data lives and when it is deleted, but the lawful basis for processing, the privacy notice, the DPIA and your answers to candidate requests are still your organisation’s work.&lt;/p&gt;

&lt;p&gt;If an ATS uses AI to screen applications, that is also a high-risk use under the EU AI Act. We explain what that means on our &lt;a href="https://vekt.eu/ai-act/" rel="noopener noreferrer"&gt;AI Act page&lt;/a&gt; and what GDPR asks of a hiring system on our &lt;a href="https://vekt.eu/gdpr/" rel="noopener noreferrer"&gt;GDPR page&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  How to choose
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;You want a dedicated ATS and a familiar workflow:&lt;/strong&gt; start with OpenCATS or FreeATS.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;You want AI screening, but with a human deciding and the data under your control:&lt;/strong&gt; Vekt, or Vellum if a shared inbox and multi-tenancy matter more to you.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;You want recruiting inside a full HR or ERP system:&lt;/strong&gt; Horilla, Frappe HR, OrangeHRM or Odoo, ideally whichever matches what you already run.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;You want a codebase to build on:&lt;/strong&gt; pick by language and licence first. MIT gives you the most freedom.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Whichever you pick, run it with real job postings for a week before you move live candidates into it.&lt;/p&gt;

&lt;h2&gt;
  
  
  FAQ
&lt;/h2&gt;

&lt;h3&gt;
  
  
  What is the best open-source ATS?
&lt;/h3&gt;

&lt;p&gt;There isn’t one best option; it depends on scope. For a dedicated ATS, OpenCATS and FreeATS are the most established. For AI screening with human review and automatic data deletion, Vekt. For recruitment inside a full HR suite, Horilla, Frappe HR, OrangeHRM or Odoo.&lt;/p&gt;

&lt;h3&gt;
  
  
  Which open-source ATS is easiest to self-host?
&lt;/h3&gt;

&lt;p&gt;The projects with a maintained Docker Compose setup: Vekt, FreeATS, Vellum and Horilla. Vekt also has a one-line setup script that generates secrets and starts everything.&lt;/p&gt;

&lt;h3&gt;
  
  
  Is there a free open-source ATS with no user limits?
&lt;/h3&gt;

&lt;p&gt;Yes. Self-hosted open-source software has no seat limits: Vekt, OpenCATS, FreeATS’s self-hosted version and the others here all let you add as many users as you like. Limits only apply to hosted plans, such as FreeATS Cloud.&lt;/p&gt;

&lt;h3&gt;
  
  
  Can an open-source ATS use AI without sending CVs to OpenAI?
&lt;/h3&gt;

&lt;p&gt;Yes, if it supports a local model. Vekt can score CVs with a model running in Ollama on your own hardware, and Vellum also supports local AI.&lt;/p&gt;

&lt;h3&gt;
  
  
  What’s the difference between a self-hosted ATS and an open-source ATS?
&lt;/h3&gt;

&lt;p&gt;Self-hosted describes where it runs: on your servers. Open source describes the licence: you can read and change the code. Most tools here are both, but a hosted service can be open source, and some proprietary software can be self-hosted.&lt;/p&gt;




&lt;p&gt;Found an error, or a project we should add? &lt;a href="https://github.com/Behnoudmst/vekt/issues" rel="noopener noreferrer"&gt;Open an issue on GitHub&lt;/a&gt; and we will update this page.&lt;/p&gt;

</description>
      <category>opensourceats</category>
      <category>selfhosted</category>
      <category>comparison</category>
    </item>
  </channel>
</rss>
