Personalization and AI

Personalization starts with your own article actions. AI-assisted views build on that corpus, but several of them are optional because a deployment operator must configure the server-side provider or storage they use.

Recommendations

Recommendations score candidate articles from local signals such as source, category, tags, discovery time, and your previous actions. Starred and Done articles are positive signals, Skip is negative, and Later is neutral. Recommended items can appear in Today with a relevance indicator and a reason.

Open Settings → Personalization and select Refresh recommendations to recompute your scores immediately. If you have little history, read, star, or skip a few articles first. Recommendation scoring itself uses data in the instance's PostgreSQL database and does not require an external AI provider.

See Recommendations for scoring and reader controls.

Reading DNA controls

Open Reading DNA from the navigation to review your recent category and source mix, average dwell time, streak, and achievements. Under Active nudges, each category slider adjusts how strongly that category contributes to your recommendation scores. The novelty slider adjusts the lift for plausible articles that differ from your starred and Done history.

Each slider ranges from 0.0× to 3.0×. Changes belong to your account, save immediately, and trigger a recommendation recalculation; an administrator does not set these weights for you. Novelty needs stored article embeddings to contribute to a score, so changing it has no effect on articles whose novelty cannot be assessed.

Ask

Ask answers questions over articles available to your account and returns cited articles that you can open. It can also recognize supported workflow requests, show a proposed action plan, and wait for your approval before changing article state.

Ask requires the deployment operator to configure a supported server-side AI provider. If it is unavailable, contact the instance operator; adding a key is not a reader setting. By default, Ask searches your Starred and Done articles and needs enough material in that corpus. Select Include all non-archived articles on the Ask page to widen the corpus when you want Today, Later, Skipped, and other non-archived articles considered too.

AI Watchlists

In Settings → AI Watchlists, describe a topic or goal, preview recent matches, give the watchlist a label, and add it. Enabled watchlists evaluate new material in the background and can create notifications for matches. They never star, archive, or otherwise triage an article for you.

Matching can use the configured AI provider when one is available and falls back to deterministic article search and scoring when it is not. Notification delivery still depends on the instance and device configuration described in Settings and account data.

AI Memory

Settings → AI Memory stores explicit preferences or goals for your account. Add a memory yourself, edit it, or deactivate it when it no longer applies. Learn from recent reading can create a small set of memories from your recent Reading DNA patterns without requiring an external model.

Active memories provide additional personal context to supported generated features such as Ask and briefings. Review them periodically; they are visible and controllable in Settings, and they are included in personal data export/restore.

Topic Map

Topic Map groups articles from the last seven days by embedding similarity. Select a cluster outline to inspect its summary and articles, or select a point to open the article.

This view needs enough recent articles with stored embeddings. Generating new embeddings is a server-side AI capability configured by the deployment operator; existing stored embeddings can still populate the map. An empty map can simply mean that the instance has not generated enough embeddings yet.

AI Stats

AI Stats offers 7-, 14-, and 30-day views:

• Word Cloud derives weighted terms locally from recent titles and summaries.
• Embedding Space projects stored article embeddings, so its coverage depends on server-generated embeddings.
• Knowledge Graph shows extracted entities and relationships. The instance can use optional graph storage and otherwise falls back to cached entity data where possible.

See Knowledge Graph for relationship meanings, provenance, and how graph context can support Ask.

Who controls what

Capability	Reader control	Instance control
Recommendations and Reading DNA	Actions, feedback, category and novelty weights, and manual refresh	Background recalculation and optional generated explanations
Ask	Question, corpus scope, and approval of proposed actions	AI provider credentials
AI Watchlists	Query, label, enabled state, and deletion	Background evaluation and notification support
AI Memory	Add, learn, edit, and deactivate memories	Availability of generated features that consume them
Topic Map and embedding views	Time range or selected cluster	AI embedding provider and processing
Knowledge Graph	Time range, filters, and selected entity	Entity extraction and optional graph storage