The News Manager Concept

How does m18t's News Manager work? Learn the four-stage pipeline — Scout, Extract, Evaluate, Generate — what's free, what needs your AI key, and item controls.

News Manager turns the RSS and Atom feeds you follow into review-ready draft articles, through a four-stage pipeline that stays scoped to a single brand.

What you'll learn

  • The mental model: feeds in, draft articles out — a pipeline, not a reader.
  • The four pipeline stages and what each one does.
  • What runs without AI versus what needs your own AI key, and what only runs on the rows you pick.

The concept: a pipeline, not a reader

A traditional feed reader collects headlines and stops there. You still have to read, judge, and rewrite everything yourself.

News Manager is built as a pipeline. It pulls articles from the sources you follow, helps you decide which ones are worth your attention, and writes a first draft you can edit. You stay in control at the end of every run — nothing is published on its own — but the repetitive collecting and summarizing happens for you.

Everything in News Manager belongs to one brand. The sources you add, the articles they bring in, and every action you take are scoped to the active brand. Switch brands and the queue re-shards to that brand's sources. Nothing crosses between brands.

You'll find it at News Manager in the Studio sidebar (/studio/news-manager), with Sources and Feed Items tabs and a pipeline toolbar across the top.

The four stages

An article moves through four stages on its way from a feed to your Content drafts. The four pipeline buttons are labelled Scout, Extract, Evaluate, and Generate.

Stage 1: Scout

Scout polls your active sources and pulls in any new articles they publish. Articles are deduplicated by their web address, so the same story is never added twice — see The Dedup Mechanism. Scout uses no AI. It can run on demand from the toolbar, or automatically on a schedule.

Stage 2: Extract

A feed usually carries only a short summary. Extract visits the original article and fetches its full body, so the later stages have the complete text to work with rather than a teaser. Extract also uses no AI.

Stage 3: Evaluate

Evaluate reads each article and scores how relevant it is to your brand, then either approves it (advancing it for drafting) or skips it with a short recorded reason. This stage uses AI and respects your brand's excluded topics. See The Evaluate Stage.

Stage 4: Generate

Generate takes an approved article and writes a draft post from it. The draft lands in your Content in a planning state, where you review, edit, and decide whether to publish. This stage uses AI, and nothing is ever published automatically. See The Generate Stage.

What runs without AI, and what costs tokens

  • Scout and Extract use no AI and cost you nothing in tokens. You can also leave scouting on a schedule.
  • Evaluate and Generate use AI, and they run only on the rows you select in the Feed Items tab. You tick the checkboxes for the articles you want, then run the step from the selection bar. The AI never works through your whole backlog on its own, so your spend stays predictable.

Bring your own AI key

Evaluate and Generate need your own AI provider key configured for the brand. Until you set one, those buttons stay disabled with a tooltip; Scout and Extract keep working regardless. This is m18t's bring-your-own-key model — your key, your bill, no markup. The brand's empty-state even spells it out: Evaluate and Generate need your own AI key; Scout and Extract work without one.

Per-item controls and recovery

Each row in the Feed Items tab has its own actions. Beyond running a stage on a single item, you can:

  • Retry an item that errored (for example, the AI returned an unreadable response).
  • Skip an item you don't want to process, which sets it aside.
  • Reset stuck items — if a run crashes mid-evaluation and leaves rows stranded in the "evaluating" state, this reclaims them so you can run them again.

These keep a long-running queue healthy without forcing you to start over.

Categories, scheduling, and housekeeping

  • Categories are your brand's own topic tags, managed inside the app. You label sources and filter incoming articles with them so a busy queue stays easy to scan.
  • Scheduling controls how often Scout runs on its own — off, or at the interval you choose — handled through a cron job behind the scenes.
  • Cleanup tidies aged and skipped feed items over time so the queue stays fast. It only touches raw feed clutter; anything you've already turned into a draft lives in your Content and is never removed by cleanup. The dedup fingerprint also survives cleanup, so cleared stories don't reappear.

FAQ

Why is the Evaluate or Generate button greyed out? Those steps need your own AI key. m18t is bring-your-own-key — add an OpenAI, Google, or Anthropic key to the brand's vault and the buttons enable. Scout and Extract work without a key. See Why m18t uses BYOK.

Does News Manager publish articles automatically? No. Generate creates a draft in your Content in a planning state. You review and publish it yourself; nothing goes live on its own.

Will the AI process my whole queue and run up a bill? No. Evaluate and Generate run only on the rows you select. You choose what gets the AI treatment.

Some items are stuck in "evaluating" — what do I do? Use Reset stuck in the toolbar. It reclaims items stranded by a crashed run so you can re-run them.

Is my news data shared across brands? No. Sources, feed items, and drafts are all scoped to the active brand.

What's next

Was this article helpful?