Settings

AI provider

Any OpenAI-compatible endpoint; the default is nano-gpt. Grouping runs often on short text; analysis runs rarely on long text — so it's usually worth a cheap model for one and a strong one for the other.

Base URLAPI key · saved

Grouping modelAnalysis model

Pipeline

Fetch full article text

Off means the analysis only ever sees feed blurbs — usually about 100 characters.

Analyse new stories automatically

Costs tokens on every cycle. Off means you press Analyse on the ones you care about.

Run every (minutes)How often the pipeline fetches and groups.Coverage window (hours)Ignore news older than this.Evidence window (hours)Primary sources stay relevant far longer than news.Story stays open (hours)How long a story keeps accepting late coverage.Minimum outletsBelow this, a cluster isn't a story worth comparing.Extract concurrencySimultaneous page fetches across all hosts.Per-host delay (ms)Minimum gap between two hits on one site.Minimum wordsBelow this, extraction counts as thin.Grouping batch sizeHeadlines sent to the grouping model at once.Chars per articleHow much of each body the analysis model sees.

Settings

AI provider

Pipeline

Recent runs