n8n Filters News Cheaper Than You'd Think: You.com, Jev, and $0.00004 Per Article

A morning news digest usually forces a choice between bad and expensive: either spam 60 articles, or spend real money running a large model across all 60, only to send a summary of those same 60 articles. People stop opening it after three weeks.

The breaking point is who pays for the filtering. The typical solution runs an expensive model across the entire volume of news (search→analyze→select). n8n's approach flips the order: search→cheap filter→expensive processing.

Architecture: Three Stages, Clean Separation

Stage 1. Search Broadly

Your n8n instance runs on a schedule every day at 8am local time. You define topics in a single Config node: three starting topics (model releases, agent tooling, AI regulation) are enough to keep the morning post readable. You.com searches for news from the last 24 hours on each topic. Stale results (from yesterday's run) are skipped — a three-day-old story doesn't end up in a fresh brief, which is what trains people to tune out the whole message.

Stage 2. The Filter: Three Questions, $0.00004

This is the heart of the workflow. Jev (System One from TypeSafe AI) isn't a chat model. You pass it a state (the article) along with a set of typed questions, and it returns typed values with calibrated confidence in half a second — no text output, and therefore no malformed-response errors.

Three questions, evaluated independently: • Is this an actual event or an echo of an announcement? (Separates a launch from commentary about it, a funding round from a roundup that merely mentions it) • Is this really about your topic? (Search drifts: a query about regulation can return a story about data-center zoning) • How material is this to your team? (Scored against the team description and priorities defined in Config — this sets the materiality threshold)

Cost: four hundredths of a cent per article. For 60 articles across three topics, that's roughly a quarter of a cent. It's precisely at this price point that filtering becomes viable; make it more expensive and you're back to the old model — analyze everything and hope to select some of it.

Stage 3. Grouping and Research

A single model release can generate six publications by noon. Six entries in a digest isn't signal — it's padding. Material articles get grouped by topic; one event covered in six places becomes a single entry with six sources. Then You.com Research writes a short, citable brief for each topic, and that goes to Slack. If a topic produced nothing material, there's no post for it at all — not a post saying there's no news.

The Numbers

A morning run costs $0.01–$0.04 (searches, Jev calls on each article, and Brief generation on topics that had news). Quiet days are cheaper: the research stage simply doesn't run for empty topics. The structural reason it's cheap: the expensive stage only ever sees what survived a filter costing four hundredths of a cent. Flip the order — run an expensive model on 60 articles, then select — and you pay for inference on 60 to get output for 4.

Three Advantages Over a Naive Digest

• Staying silent on quiet days is the key mechanism that keeps people opening this in month three • A materiality threshold you simply adjust in one node, instead of re-negotiating in a prompt every time • Citation on every fact — the brief is forwardable, it's not rumor, it's a sourced claim

How to Get Started

You'll need: n8n (cloud or self-hosted), You.com nodes (via Community Nodes), a You.com API key, a TypeSafe key (Jev is in early access), and a Slack channel. Import the workflow template — all nodes are already wired and configured. Edit the Config: your topics, your team description, and what you care about (a competitor's release, a pricing change, regulation with a compliance date). Run it, then tune the materiality threshold after a week.

From there, you can: run heavy analysis in the background on big days (async Research with polling), add a fourth question (Jev computes it in parallel, barely changing response time — for example, urgency, which routes a critical topic into an urgent message), restrict sources (include_domains, exclude_domains, boost_domains in You.com, up to 500 values), or archive full articles via Contents. You can also swap the output destination — email, Notion, a spreadsheet, anything instead of Slack.