Most people rely on a few feeds or social timelines and miss the context that turns scattered posts into a coherent story. Horizon treats news curation as a pipeline: fetch → deduplicate → AI-score → enrich → summarize → deliver. That flow preserves human-curated sources while using LLMs to reduce noise, surface discussion, and add background — then publishes a compact daily briefing in English and Chinese.
What Sets It Apart
- Multi-source pipeline with deduplication: merges the same story across HN, Reddit, RSS, Telegram, Twitter/X and GitHub so your briefing shows the consolidated narrative rather than duplicate links — which reduces redundancy in daily digests.
- Pluggable multi-provider scoring and enrichment: supports OpenAI-compatible APIs plus Claude, Gemini and local providers; you can tune thresholds per source so the system filters by your taste rather than a fixed heuristic.
- Bilingual output and delivery flexibility: generates English and Chinese summaries from the same pipeline and can publish to GitHub Pages, send email newsletters, or push to chat/webhook endpoints for automation.
- Community + human-in-the-loop design: keeps the human layer (shared source hub, comment collection and community-sourced sources) instead of fully autocratic summarization, making it easier to capture human judgment and niche feeds.
Who It's For & Tradeoffs
Great fit if you want a self-hosted, configurable daily news radar that blends human-curated sources with LLM-powered scoring and context. Ideal for engineering teams, researchers, or individuals who track technical communities and need bilingual briefings.
Look elsewhere if you need a turnkey SaaS with zero maintenance, minute-by-minute breaking-alert latency, or a fully hosted managed newsletter — Horizon requires configuration (API keys, sources), some ops (Docker/GitHub Actions), and will incur LLM API costs for scoring and enrichment. Its quality depends on your source choices and model prompts, so expect tuning during early runs.
