Thysrael/Horizon

Ranked Daily Briefing

Context, Summary & Discussion

More Screenshots

Terminal Output	Feishu Notification
Email Delivery	WeChat Delivery

Good reads are scattered across feeds, forums, and timelines. Your attention isn't unlimited. Horizon collects, filters, and deduplicates them, then brings the worthwhile ones to you with context and community discussion.

Your taste shapes what you read—and what you want from it. A news story calls for “why it matters”; an engineering deep dive calls for “what I can use.” Horizon's profiles give each kind of content its own scoring criteria and output, for a briefing that feels handpicked.

• 📡 Bring Your Sources Together — Follow RSS, Hacker News, Reddit, Telegram, X, GitHub, financial news, and more.
• 🎯 Decide What's Worth Reading — Define scoring rubrics with profiles, set individual thresholds, and merge repeated stories within each profile.
• 🧩 Give Every Read the Right Treatment — Choose summaries, context, solutions, or takeaways with Markdown prompts and JSON block definitions.
• 💬 Go Beyond the Headline — Add web-researched background and community discussion where they help explain the story.
• ⚖️ Make Room for All Your Interests — Cap the briefing's length and each category's share so one busy topic doesn't crowd out the rest.
• 📬 Read Where You Already Are — Generate English and Chinese briefings as Markdown, publish to Pages, or deliver by email and webhook.

A profile is a reusable set of editorial rules: what belongs, what's worth keeping, and what to write. Built-in examples:

Assign a profile to a source, or let AI choose. Want a different take? Adapt an existing profile—usually without changing Python. Make a profile your own →

Editable OmniGraffle source

Profiles define the treatment; runtime settings reflect your reading preferences. Each item routes to one profile. Analysis, filtering, and deduplication happen before enrichment, then the selected items become a briefing grouped by profile.

Run the full pipeline through the CLI, or let an AI assistant call its stages through MCP.

Option A: Local Installation

git clone https://github.com/Thysrael/Horizon.git
cd Horizon

# Install with uv (recommended)
uv sync

# Install test/development extras when needed
uv sync --extra dev

# Or with pip
pip install -e .

dev is currently defined as an optional extra in pyproject.toml, so use uv sync --extra dev for pytest and other development dependencies.

If you want the optional OpenBB financial-news source, install its extra too:

uv sync --extra openbb

If openbb pulls packages without wheels on your machine, install the SDK manually with binaries only:

uv pip install --only-binary=:all: openbb openbb-benzinga

Option B: Docker

git clone https://github.com/Thysrael/Horizon.git
cd Horizon

# Optional: build with comma-separated extras before the first run
docker compose build --build-arg EXTRAS=openbb horizon

Full-text extraction with trafilatura is included in the base install. The twitter extra also requires a Playwright browser and system packages, which the current Dockerfile does not install.

Option A: Interactive wizard (recommended)

uv run horizon-wizard

The wizard asks about your interests (e.g. "LLM inference", "嵌入式", "web security") and auto-generates data/config.json. See Interactive Wizard for CLI options.

Option B: Manual configuration

cp .env.example .env          # Add your API keys
cp data/config.example.json data/config.json  # Customize your sources

Minimal manual configuration:

{
  "ai": {
    "provider": "openai",
    "model": "gpt-4",
    "api_key_env": "OPENAI_API_KEY"
  },
  "sources": {
    "rss": [
      {
        "name": "Simon Willison",
        "url": "https://simonwillison.net/atom/everything/",
        "profile": "tech-news"
      }
    ]
  },
  "processing": {
    "profiles_dir": "profiles",
    "default_profile": "tech-news",
    "profile_settings": {
      "tech-news": {
        "threshold": 7.0,
        "topic_dedup": true
      }
    }
  }
}

An explicit source profile uses that profile directly. Omit it or set it to "auto" to let AI match the item against all available profiles. Set it to an array such as ["tech-news", "finance-news"] to restrict AI matching to those profiles. See Processing Profiles for profile structure and behavior. Per-profile user preferences such as score thresholds and topic deduplication belong in processing.profile_settings, not in the profile files.

Balanced digest (optional)

Limit the final digest size and prevent one category from dominating the results. Categories come from source configuration such as sources.rss[].category.

{
  "digest": {
    "max_items": 20,
    "category_groups": {
      "ai": {
        "limit": 5,
        "categories": ["ai-news", "ai-tools", "machine-learning"]
      },
      "finance": {
        "limit": 5,
        "categories": ["finance", "business", "equities"]
      }
    },
    "default_group": "other",
    "default_group_limit": 3
  }
}

Group limits are applied after profile filtering and before enrichment. If category_groups and max_items are omitted, no balanced digest limits apply.

api_key_env must be the name of an environment variable, not the API key itself. Put the real secret in .env:

OPENAI_API_KEY=sk-your-key

For Gemini, use GOOGLE_API_KEY:

{
  "ai": {
    "provider": "gemini",
    "model": "gemini-2.0-flash",
    "api_key_env": "GOOGLE_API_KEY"
  }
}

Any string value in data/config.json can reference environment variables with ${VAR_NAME}. This is useful for values such as ai.base_url, private RSS feed URLs, webhook endpoints, or custom header templates.

For the full reference, see the Configuration Guide.

A. Local installation

uv run horizon [OPTIONS]

B. Docker

docker compose run --rm horizon [OPTIONS]

--data-dir changes the state directory, including summaries, subscribers, and the default config location; --config changes only the config file. The generated report is saved to data/summaries/ (or <data-dir>/summaries/ if --data-dir is set). See Configuration Paths for combining both flags and initializing a custom config location.

Schedule Horizon with GitHub Actions using the daily workflow template. The template is disabled in this repository; configure it for your deployment and rename it to daily-summary.yml to enable it.

Horizon can publish or deliver the generated briefing in several ways:

For delivery setup, see the Configuration Guide. To run pipeline stages from an AI assistant, use the MCP Server: tools · client setup.

Horizon is an open-source project maintained in spare time. If you'd like to support the project or be listed here, feel free to open an issue or email me.

Supporter	Details
	Thanks to Modelflare for sponsoring this project! Modelflare: Full strength, stable, nothing watered down. Global SOTA models at a lower cost. Operated by US-based Havenbyte LLC, Modelflare supports GPT, Claude, Gemini, Grok, and other leading models through common compatible protocols, plus 40+ image and video models. Cache compensation is a highlight: selected OpenAI groups (price-first, stable, and premium) guarantee daily cache hit rates of 65%, 75%, or 85%, and if eligible requests fall short, Modelflare refunds the difference, visible on the usage dashboard after settlement. Special offers: a single top-up of US$20 unlocks the GPT discount group at a 0.015x price multiplier, and a single top-up of US$50 unlocks the Claude discount group at a 0.03x multiplier. Register via this exclusive link to get started.
	Compshare currently supports Horizon. Compshare is UCloud's AI cloud platform, offering cost-effective monthly and pay-as-you-go domestic model agent plans starting from RMB 49/month, as well as stable officially relayed overseas models. It supports Claude Code, Codex, and API usage, with enterprise-grade high concurrency, 24/7 technical support, and self-service invoicing. Register through their link to receive a free RMB 5 trial credit.
	OfoxAI is a unified AI API platform bringing together text, image, and video models from multiple providers. With OpenAI-compatible endpoints and native Anthropic and Gemini interfaces, developers can access models for AI applications, agents, and content creation through one platform. Explore OfoxAI models and APIs →

Horizon already supports the full daily briefing loop: multi-source collection, profile-driven analysis and enrichment, deduplication, comment summaries, bilingual generation, GitHub Pages publishing, email delivery, webhook delivery, WeChat delivery, Docker deployment, MCP integration, and the setup wizard.

Planned improvements:

• More source types, such as Discord
• Publish releases on GitHub
• Publish the package to PyPI for pip install

Join the Horizon user and developer QQ group to share sources, profiles, and deployment tips.

QQ group: 1106121909
Scan with QQ or search the group number to join.

Contributions are welcome. See CONTRIBUTING.md for code, documentation, and source-sharing guidelines.

Want to share valuable source discoveries with the Horizon community? Please submit them through horizon1123.top.

• Special thanks to LINUX.DO for providing a promotion platform.
• Special thanks to HelloGitHub for valuable guidance and suggestions.
• Special thanks to AIGC Link for the promotions on XiaoHongShu.

MIT