GitHub Trending: Today — Authentik, HRMS, Plane, changedetection.io, Khoj
Today’s non-cooldown list is a self-hosting toolkit: identity, people operations, project work, website monitoring, and a personal AI layer. The pattern is not another developer framework; it is software that gives small teams an alternative to a service bundle.
In today’s GitHub Trending: Authentik (self-hosted identity), HRMS (HR and payroll), Plane (project management), changedetection.io (website-change alerts), and Khoj (an AI assistant over your own data).
The common thread is operational ownership. Each project replaces a SaaS surface, but the active issue queues show that self-hosting exchanges subscription simplicity for work that must be maintained.
The self-hosted office stack
| Repo | What it does | Today’s growth |
|---|---|---|
| goauthentik/authentik | Identity and SSO platform | +138 stars |
| frappe/hrms | HR and payroll software | +49 stars |
| makeplane/plane | Project-management workspace | +95 stars |
| dgtlmoon/changedetection.io | Website-change monitoring | +176 stars |
| khoj-ai/khoj | Self-hostable AI assistant | +81 stars |
This is a practical cluster rather than a single technical wave. The five repositories cover the operational surfaces that usually end up split across hosted vendors: authentication, staffing, project planning, monitoring, and search or AI assistance. Their popularity says less about a universal anti-SaaS turn than about teams seeking control where a default tool has become expensive, rigid, or too opaque.
goauthentik/authentik — identity without the identity vendor
Authentik is for teams operating their own applications and infrastructure that need SSO, OAuth/OIDC, SAML, proxies, and access control in one system. A recent community post highlighted release-candidate fixes for administration forms and proxy-provider redirects—exactly the unglamorous paths on which identity software earns trust. Its issue queue also shows the cost of owning that layer: upgrade and forward-auth edge cases are work someone must diagnose.
makeplane/plane — a project workspace teams can run

Plane targets teams that want issues, sprints, docs, and triage without accepting the limits of a hosted Jira, Linear, or ClickUp account. HN’s first question was the correct one: how does it compare with Linear? The answer matters more than its star count, because a migration only makes sense when workflow and keyboard habits carry over. Its active SMTP issue is a reminder that collaboration software becomes infrastructure as soon as it must reliably send mail.
dgtlmoon/changedetection.io — watch pages that do not notify you
changedetection.io is for people who need to know when a web page changes: price trackers, availability checks, document monitors, and researchers watching public sites. One HN commenter gave the clearest use case: government contracting pages can hide consequential new PDFs without a usable subscription feed. Recent issues around cached responses and XPath filters show that the hard part is not noticing a changed page; it is doing so consistently across the strange behavior of the web.
khoj-ai/khoj — an assistant pointed at your own corpus
Khoj is for users who want an AI interface over their documents while retaining a self-hosted option and choice of local or cloud models. Its community descriptions emphasize Markdown, PDFs, web search, Obsidian, WhatsApp, and custom agents. The issue list points toward connectors and tools, including a request for Logseq support: the useful competition is not merely model quality, but how naturally an assistant reaches the material people already keep.
The rule that emerges is simple: self-hosting is no longer a single category. It is a bundle of replaceable SaaS boundaries. The teams best positioned to benefit will not self-host everything; they will choose the boundary where control, privacy, and integration are worth the operational responsibility.
Get breakdowns like this every day
The top Hacker News stories, unpacked — by email, Telegram, or RSS.
AMD buys Taalas before its model-in-silicon hardware can ship
AMD is buying an inference startup whose wager is unusually literal: put a trained model into the chip instead of repeatedly pulling its weights from memory.