OpenReader

The problem

Feedly and Inoreader charge monthly for things that are really just config on a server: regex filtering, folders, full-text extraction, newsletter ingestion and AI summaries, which they resell at a premium. None of it needs to live behind someone else's paywall, and none of it needs your reading history in a third party's database.

The solution

OpenReader is a self-hosted RSS, Atom and email-newsletter reader that puts every one of those features back under your control, with your data in a SQLite file you own. It runs daily on a $5/month 1 GB VM, sharing the box with a full WordPress install.

Features

• RSS and Atom feeds with conditional GET, so a routine refresh costs almost nothing.
• Email newsletters over IMAP, read-only, with an app password and no OAuth.
• Content filtering with per-source regex rules. Every kept article records which rule let it in, so filtering is auditable.
• Full article mode that extracts the full page the first time you open a truncated item, then caches it.
• Summarize, faithful to the article's own text, running on your own Claude subscription instead of metered API billing.
• No scheduler anywhere. Refresh is a button; so is summarize.

How it's built

• Backend: Python and FastAPI with a synchronous refresh loop.
• Storage: one SQLite file.
• Newsletters: sources sharing a mailbox folder are batched into one IMAP pass, over one reused connection.
• Summaries: shells out to the claude CLI on demand, behind a hard kill switch.
• Frontend: React, a three-pane shell with keyboard navigation.

The hardest part

Making it comfortable on a 1 GB VM shared with WordPress. The first real summarize call was OOM-killed at a 300 MB ceiling, because the CLI's peak memory was far above the app's baseline. The fix was measuring the real subprocess footprint rather than guessing.