Horizon

Open source alternative to Feedly

12.5K stars
MIT
Active this week
Last commit 1 day ago

Last verified Nov 5, 2026

Automate daily news briefings from RSS, Hacker News, Reddit, and more, scored and summarized by AI profiles, then delivered to email or webhooks.

Visit website GitHub12.5K Deploy on Hostinger

Read the full breakdownWho it is for, how it works, trade-offs, the quick start and the FAQ

who it's for

Who Horizon is for#

Engineers tracking fast-moving technical topics

Configure RSS feeds from your preferred engineering blogs, Hacker News top stories, and relevant subreddits into a tech-news profile. Set a scoring threshold and topic deduplication to get one ranked briefing per day covering what actually matters across all those sources. Delivery to email or a webhook means the briefing arrives where you already work.

Skip if:

You follow fewer than five sources and read everything manually without backlog. At that volume, a plain RSS reader requires less setup and no LLM API key.

Researchers monitoring AI and ML developments

Point Horizon at ML conference proceedings feeds, ArXiv RSS, AI lab blogs, and curated subreddits with a custom profile that asks for practical takeaways and new technique summaries. The LLM scoring ranks items by relevance to your defined criteria, not by recency or upvotes.

Skip if:

Your research is confined to one or two authoritative sources. A single RSS subscription is less setup and has no ongoing API costs.

Teams delivering a shared internal news digest

Assign sources to multiple profiles based on content type (AI news, finance, engineering), set category limits in the digest config, and deliver a single balanced briefing to email or a team webhook. Each profile independently scores and formats its section, so an AI-news entry reads differently from a finance item.

Skip if:

Your team needs a no-infrastructure option with a GUI. Horizon requires a server or scheduled cloud environment and is entirely config-file and CLI driven.

Developers building AI content workflows

Wire Horizon's MCP server into your AI assistant and use natural language to configure sources, adjust scoring profiles, and trigger runs. The agent skill installation makes this accessible without editing config files directly. Outputs go to webhooks that plug into downstream pipelines.

Skip if:

Your workflow needs a visual dashboard or mobile app. Horizon produces briefings as files, emails, and webhook payloads; there is no built-in web UI.

the problem

The problem it solves#

Keeping up with tech news, engineering blogs, and community discussions across a dozen channels is a real time sink. RSS readers accumulate thousands of unread items. Hacker News top stories repeat across feeds. Important threads on Reddit get buried under noise. Paid readers like Feedly offer AI-assisted filtering, but the scoring criteria are fixed by the product, not configurable per source or interest type.

The harder problem is signal quality. A technology briefing and an engineering deep dive need different treatments. News calls for impact and community context; a blog post calls for the practical takeaway you can use. No commercial reader lets you define those criteria per source type. Teams and individuals who want a genuinely tuned daily briefing often end up writing ad-hoc scripts, which break with API changes and have no path to multi-channel delivery.

how Horizon solves it

How it solves it#

Multi-source aggregation

Pulls from RSS and Atom feeds, Hacker News, Reddit, Telegram, X, GitHub stars, and financial news (via OpenBB) in a single pipeline run. Any public RSS or Atom feed is supported; Telegram channels, X accounts, and subreddits need their respective credentials. Each source is assigned to a profile, or Horizon auto-routes it via LLM matching.

Profile-based AI scoring

Each profile defines a scoring rubric, a minimum threshold (e.g., 7.0), and a topic deduplication setting. A tech-news profile scores for impact and community discussion; a tech-blog profile looks for practical takeaways. Sources inherit a profile explicitly, or the LLM assigns one automatically. Thresholds are configured per profile in processing.profile_settings without touching Python.

Configurable output templates

Markdown prompts and JSON block definitions control what each briefing section contains: a headline summary, background context, a recommended action, a takeaway, or community discussion links. The structure is per-profile, so an AI-creator briefing looks different from a tech-news digest. Adjust what gets written by editing the profile; no Python changes needed for most customizations.

Multi-channel delivery

Output reaches Markdown files, GitHub Pages, email (SMTP), Feishu bots, WeChat, and any webhook endpoint. Each channel is configured independently; a single pipeline run can publish to multiple targets simultaneously. English and Chinese briefings can be generated in the same run.

Balanced digest limits

Category groups (such as ai, finance, or other) each carry a configurable limit so one busy topic does not crowd out the rest. digest.max_items caps the total; category_groups controls the per-category share. These limits apply after AI scoring and before enrichment, so only relevant items compete for the final slots.

MCP and AI assistant integration

An MCP server lets any compatible AI assistant configure sources, adjust profiles, and trigger pipeline runs directly through natural language. The agent skill installs with `npx skills add Thysrael/Horizon --skill horizon -g` and requires Node.js 22.20 or later.

strengths · trade-offs

Strengths and trade-offs#

Strengths

MIT license with no subscription costRun the full pipeline on your own server indefinitely for free. MIT lets you use Horizon commercially and in team settings with no attribution fee beyond keeping the license notice. Unlike Feedly's paid AI plans, you pay only for LLM API calls and your own compute.
LLM-agnostic pipelineWorks with OpenAI GPT models, Anthropic Claude, Google Gemini, DeepSeek, Bytedance Doubao, MiniMax, and local models through Ollama. Swap the provider by changing three lines in data/config.json. ChatGPT subscribers can use the OpenAI ChatGPT integration without a separate API key.
Stable core with 12,652 stars and 3 open issuesHorizon reached 12,652 stars and 1,829 forks with only 3 open issues as of 2026-11-05. The low open issue count relative to the repo size suggests the core pipeline is stable and the maintainer is responsive. The repo was created in February 2026 and last updated November 2026.
Built-in GitHub Actions schedulingA daily workflow template ships in .github/workflows/daily-summary.yml.disabled. Rename it to daily-summary.yml to enable unattended daily runs with no persistent server infrastructure beyond the repo itself and your LLM API key.

Trade-offs

-No prebuilt package or Docker imageBoth installation paths (local and Docker) start with git clone. There is no published PyPI package, no Homebrew cask, and no prebuilt Docker image. Setup requires cloning the repository and running uv sync or docker compose build. This is a moderate barrier for users who expect a one-line install.
-LLM API costs scale with source volumeEvery scored item calls the configured LLM. High-frequency sources with many daily items can generate meaningful API spend, particularly with GPT-4 or Claude-tier models. The scoring threshold and max_items cap help control volume, but there is no built-in token budget guard.
-Twitter/X source excluded from DockerThe README states that the twitter extra requires Playwright and system packages not included in the current Dockerfile. Running Horizon with X as a source requires a local install and manual Playwright setup outside of Docker.

versus alternatives

Horizon vs alternatives#

Horizon vs Feedly

Feedly and Horizon both aggregate news from multiple RSS and social sources into a readable briefing. Feedly is a managed subscription service with AI features on paid tiers; Horizon is a self-hosted Python pipeline where you configure the LLM, the scoring criteria, and the delivery channel.

Feature	Horizon	Feedly
License	MIT	Proprietary
Self-hosting	Yes	No
LLM provider	Your choice (OpenAI, Claude, Gemini, Ollama, more)	Feedly AI (Leo)
Sources	RSS, HN, Reddit, Telegram, X, GitHub, finance	RSS, Twitter, newsletters
Scoring control	Custom rubrics, thresholds, per-profile settings	Fixed topic priorities
Delivery	Email, webhooks, WeChat, Feishu, GitHub Pages	In-app, email digest, Slack/Teams
Setup	Clone, configure, schedule	Sign up, add sources

Horizon is the stronger choice when scoring transparency and control matter. If you need a briefing tuned to specific criteria without a subscription, Horizon's profile system handles that directly. The pipeline also delivers to channels Feedly does not support, including WeChat and arbitrary webhooks.

Feedly is still the better fit when setup time is a constraint. Feedly's web app requires no server, no API keys, and no config files; sources are added through a UI in minutes. For individuals or small teams who want news aggregation without any infrastructure, Feedly's free tier covers basic use and its paid AI tier adds filtering without self-hosting.

install · quick start

Quick start#

bash

Install Horizon by cloning the repository and syncing Python dependencies with uv.

```bash
git clone https://github.com/Thysrael/Horizon.git
cd Horizon
uv sync
uv run horizon
```

tech stack · detected from GitHub

What it's built on#

Languages: Python

frequently asked

FAQ#

Does Horizon require a paid LLM API subscription?

Yes, Horizon calls a configured LLM to score and summarize items. It supports OpenAI (GPT), Anthropic (Claude), Google (Gemini), DeepSeek, Bytedance Doubao, MiniMax, and local models via Ollama. ChatGPT subscribers can use the OpenAI ChatGPT integration without a separate API key. If you use Ollama with a local model, there is no per-call cost beyond your own compute.

Can I run Horizon on a schedule without keeping a server running?

Yes. The repository includes a GitHub Actions workflow template (.github/workflows/daily-summary.yml.disabled) for daily automated runs. Rename it to daily-summary.yml and configure your secrets to run Horizon on a schedule with no persistent server. The generated report is saved to the data/summaries directory and can be delivered to email or a webhook from within the same workflow.

How do I add a new source to Horizon?

Add an entry to the sources section of data/config.json with the source type, URL or handle, and an optional profile assignment. For RSS, that is a URL and a name; for Reddit, a subreddit name. The interactive wizard (uv run horizon-wizard) generates the config file from your interests if you prefer not to edit JSON by hand. Omit the profile field to let Horizon auto-route the source.

What does Horizon do with duplicate stories across sources?

Within each profile, setting topic_dedup to true merges repeated stories before scoring. Items that share a topic (detected by the LLM) are grouped and counted once in the final briefing. This prevents a major story from occupying multiple slots when it appears across HN, Reddit, and several RSS feeds simultaneously.

Is Horizon available as a hosted service?

No. Horizon is a self-hosted pipeline only. There is no managed cloud version; you run it on your own server, laptop, or in a GitHub Actions workflow. The website at horizon1123.top is a project landing page and community hub, not a hosted service.

also worth a look

Similar open-source tools#

Miniflux

Alternative toFeedly

Self-hosted feed reader with tracker removal and 25+ integrations.

9,890 GitHub stars
Licence Apache-2.0

FreshRSS

Alternative toFeedly

Self-hosted RSS aggregator for news, podcasts, and video channels.

16,495 GitHub stars
Licence AGPL-3.0

smallweb

Alternative toFeedly

Discover personal blogs and indie creators, not algorithms

1,728 GitHub stars
Licence MIT

OpenClaw

Featured

Alternative toMuse

Self-hosted AI assistant for WhatsApp, Telegram, and more

393,917 GitHub stars
Licence MIT

Inbox Zero

Featured

Alternative toSuperhuman

AI email assistant that drafts replies and clears your inbox

12,889 GitHub stars
Licence AGPL-3.0

rakazo

Featured

Alternative toGrok Bot

AI teammates you own: your keys, your model, your machine.

7,028 GitHub stars
Licence Apache-2.0

by category

Open source tools in these categories#

481

AI & Machine Learning

freeCodeCamp, REA (Reverse Engineer Anything), OpenClaw 275

Business & Productivity

OpenClaw, Excalidraw, RustDesk

Hostinger

Self-host Horizon on a Hostinger VPS

Keep it running 24/7, even when your laptop is off. A Hostinger VPS gives it a server of its own.

01
Choose a plan
Starts from $6.49/mo
02
Apply
Extra 10% off yearly plans
03
Install Horizon
On a server that stays on
04
Start building
Focus on the app, not the server

Deploy on Hostinger

Inbox Zero

Alternative toSuperhuman

AI email assistant that drafts replies and clears your inbox

12,889 GitHub stars
Licence AGPL-3.0
#Email

rakazo

Alternative toGrok Bot

AI teammates you own: your keys, your model, your machine.

7,028 GitHub stars
Licence Apache-2.0
#AI Agents

SnitchFeed

Live monitoring on LinkedIn, X, Reddit & more. Agentic social listening for startups.

Marketing & Growth

Advertise your product

The projects we add and every new guide. Free, unsubscribe any time.

Find open source alternatives to paid software

Horizon vs Feedly

Feature	Horizon	Feedly
License	MIT	Proprietary
Self-hosting	Yes	No
LLM provider	Your choice (OpenAI, Claude, Gemini, Ollama, more)	Feedly AI (Leo)
Sources	RSS, HN, Reddit, Telegram, X, GitHub, finance	RSS, Twitter, newsletters
Scoring control	Custom rubrics, thresholds, per-profile settings	Fixed topic priorities
Delivery	Email, webhooks, WeChat, Feishu, GitHub Pages	In-app, email digest, Slack/Teams
Setup	Clone, configure, schedule	Sign up, add sources