Introduction
—「Taste it. Savor it.」
Reading is like sipping soup — don't keep glancing into the bowl, close your eyes and savor it first.
sip is an information firewall: letting you and your AI only see content you trust.
It is not an algorithm-driven reader, nor a feed meant to make you "scroll more". It is a local-first transparent information filter and reading assistant — you specify your sources, sip guards and assists in filtering them while improving the reading experience, so you and your AI agents get answers from a clean, traceable dataset.
And, shamelessly: https://blog.hotsouprealm.top/atom.xml Follow the Hot Soup Teahouse. Follow us, thank you. 🐾
Why does sip exist?
Today's information environment has three harsh realities:
- LLMs cite garbage sources — DeepSeek, Doubao, ChatGPT casually cite Sohu, Baijiahao, low-quality self-media, and you can't even tell it "don't use these".
- Algorithms trap you in a bubble — TikTok/Toutiao keep you scrolling but never tell you "why this was recommended", and never let you choose "what I don't want to see".
- The people you care about lack the ability to discern — your parents, your friends have no technical means to protect themselves against clickbait and rumors.
sip's answer to these three problems is simple:
I stand my ground, and I read the information too.
Core Design Principles
| Principle | Description |
|---|---|
| Local-first | Data stays in your hands (SQLite + file cache), no account needed, reading history never uploaded |
| Transparent decisions | Only see sources you subscribe to, no algorithmic black box; filter rules are simply the feed list you maintain |
| Version is truth | What did the author change? When? sip records it all for you, losing no history |
| AI reads a whitelist only | AI summaries and semantic search rely only on sources you trust, eliminating low-quality citations |
| Out of the box, ultra-light | Single-file exe, zero dependencies, works on launch; AI features called on demand, never pre-run |
What specific problems does it solve?
1. A guardrail for AI
When you or your AI agent need to look something up:
- Let AI call
sip --search "xxx" --json - AI only retrieves from the sources you subscribe to
- Say goodbye to "AI rambling on about Sohu and Baijiahao"
2. Make information changes "visible"
A normal RSS reader only tells you "there's an article".
sip tells you:
- "This article was modified by its author on August 1"
- "This is what it said before, and this is what it says now" (
sip --diff 123 v1 v3) - "This blog changed its key opinions 12 times over the past year"
What you see is no longer a static page, but the evolution of information.
3. Help the people you care about avoid information overload
Configure a whitelist for your parents (for example: CCTV News, your local weather bureau, medical accounts you trust).
After they open sip:
- They only see the sources you've curated
- Articles with too-short summaries are automatically prompted to fetch the full text, to be read slowly
- No need to tell truth from falsehood, because the garbage sources are already blocked out
hahahotsoup note: I'm well aware the TUI has a high learning curve, so once the program matures, Avalonia is on the roadmap.
Design Boundaries
sip only does two things — deterministic rules and local fact storage; anything requiring judgment/explanation/conversation is left to agents and users. Today's selection rules for the Hot Soup Teahouse remain fixed and explainable; once enough telemetry data accumulates for personalization and other "intelligence", it will either be implemented as new deterministic rules or live in the agent layer — not in the program.
In Closing
sip won't make you scroll more, but it will let you read with more peace of mind.
May we meet again in good health 🍲