Eigenentwicklung

If you want to stay current professionally, you subscribe to newsletters. If you're serious, several — industry news, product updates, technical deep-dives, tools-of-the-week. After a few weeks the inbox is a graveyard: unread, unsorted, eventually deleted. The content was valuable; the form makes it unusable. Layout tables, tracking pixels, "click here" buttons, triple-nested footers with unsubscribe links — and somewhere in the middle, the paragraph that was the actual point.

Honestly: who really searches their newsletter archive in the mail client two years later?

A command-line tool that collects newsletter emails, strips the HTML clutter, summarises with AI, optionally translates, and files them as clean Obsidian notes in your own knowledge base — complete with tags, short summary, and source reference.

The flow

1. Fetch the newsletter — the application connects to the mail server (IMAP, POP3, or Exchange On-Premises) and reads new emails. Which senders are eligible at all is configured; already-processed emails are reliably skipped.
2. Separate signal from noise — not every email from a newsletter sender is content. A two-stage classification — first rule-based, then a cheap AI model — recognises ads, dispatch confirmations, and order notifications and filters them out before expensive processing starts.
3. Clean up HTML — a five-stage pipeline removes tracking pixels, ad banners, layout tables, and footer remnants. It detects the dispatch platform (Mailchimp, Substack, beehiiv, and friends) and extracts the actual content — data tables are deliberately kept, layout tables are dissolved.
4. Take the images along — embedded graphics and externally referenced images are downloaded in parallel, placed in a daily folder, and linked correctly in the markdown. Even if the sender shuts down their image CDN in two years — the note stays complete.
5. HTML becomes Markdown — a tailored conversion produces Obsidian-compliant Markdown with ATX headings, properly formatted data tables, and clean links — no HTML residue, no broken lists.
6. Understand the content — an AI model reads the newsletter and returns a short summary (TLDR), matching tags, and a meaningful title — structured, not as free text.
7. Translate where needed — English newsletters are translated to German. A placeholder mechanism reliably protects URLs, image references, and code blocks from being "co-translated".
8. File in Obsidian — every note gets structured metadata (YAML frontmatter), a TLDR box up front, the prepared content, and a reference to the source email. Filenames follow a configurable pattern; German translations sit next to the original with a _de suffix.

Multiple models, one codebase

Which AI model does the work is a configuration question — OpenAI, Anthropic, Azure OpenAI, OpenRouter, or a local model running on Ollama. Different models can be chosen for the cheap classification and the more demanding content analysis; per run, which model was used when is logged.

Delivery as a self-contained application

The tool ships as a standalone program (Windows or macOS) that runs without a separate Python install. Configuration and credentials live in two files alongside it — what can be maintained on site can be maintained on site.

• Inbox clutter becomes a knowledge base. What used to gather dust in the mailbox and eventually got deleted is now searchable, linkable, and part of your own research.
• Tracking pixels and ad footers end up where they belong — in the bin. What's left is the content.
• English isn't a barrier any more. Read the original if you prefer English; have the German translation alongside if you don't — at the press of a key.
• Images stay available offline. Content is still fully readable in two years, regardless of whether the sender still hosts the newsletter.
• Tags emerge automatically and consistently. What used to be tedious manual work sorts itself — and makes the archive actually navigable.

• The choice of sources. Which senders are processed at all is decided by a person — once, in the configuration. That way the whole inbox doesn't drift into the knowledge archive, only what belongs there.
• The reading, linking, commenting. The tool prepares — reading, connecting to your own notes, condensing into cross-references happens in Obsidian, as before.
• The inbox. Local Markdown files are produced. No cloud sync of email content, no extra data flows — the inbox stays the inbox.

The setup works wherever HTML, PDF, or web content from emails or feeds regularly needs to flow into a structured knowledge or document base — and the content-side handling stays with a person: tender mailings, supplier and product updates, compliance and regulatory bulletins, market-research briefings, industry press digests, competitor monitoring.

The pattern: source → automatic clean-up → AI understanding → structured filing in your own knowledge base.

AI turns noise into content. The filing structure and the reading workflow stay where they belong — with the person who actually uses the content.