Read thirty sources. Publish one piece.
The four-step pipeline you'd otherwise do by hand: scout RSS sources, extract article bodies, AI-evaluate relevance, AI-draft a publishable piece. Deduplication built in. Drafts land in your content workspace ready for editorial review.
The same four steps. Different jobs.
Niche content creator
You publish daily or weekly about one topic. You read 20-30 RSS sources to find what to write about. m18t scouts, extracts, evaluates, drafts — you do final edits and publish. Three hours of reading-and-deciding compressed into thirty minutes of review.
Small news team
You run a vertical news site. Multiple writers, regular publication cadence. m18t handles the source-monitoring and first-draft generation; your editors take it from there. The pipeline reduces drafting time, not editorial judgment.
Competitive intelligence for marketers
You're a marketer watching what competitors publish. Set up their blogs / news feeds as sources, run the Scout step (skip Evaluate / Generate). The feed-items list becomes your weekly intelligence dashboard.
Four stages. Each independently runnable.
- 1
Scout
Poll all active RSS sources for new articles. Per-source poll interval (default 12 hours, adjustable). Dedup via URL hash so the same article never gets ingested twice. New feed items land with status "discovered."
- 2
Extract
Fetch full article body for discovered items (some RSS feeds only provide summaries). Stored on the feed item. Optional — competitive watchers might skip this.
- 3
Evaluate (AI, BYOK)
Run an AI relevance check against each feed item. The evaluation prompt is in your prompt library, editable. Output: relevance score + decision. Irrelevant items move to "skipped" with a reason; relevant items move to "approved."
- 4
Generate (AI, BYOK)
For approved items, run an AI drafting step. Output: a new content row in your content workspace, with the source URL and source name populated. Status: typically "doing" or "testing" — your editorial workflow takes over from there.
Where the news pipeline stands.
Solid today
- — End-to-end pipeline runs in-app
- — URL dedup via hash
- — Per-source poll interval, last polled, last error
- — Brand × category × region × language scoping
- — Drafts link back to source feed item
Rough today
- — Generate step wired for one specific niche / locale
- — Source-discovery (suggesting new feeds) is manual
Coming
- — Locale-and-topic-generic Generate
- — Per-source AI evaluation prompts
- — Suggested sources based on existing tags