🚀 Alif-Digest

The Autonomous AI Signal Digest CLI.

Alif (ألف) scours the web for high-signal AI developments, cuts through the noise using a multi-layer scoring system, and delivers a curated daily digest directly to your workspace — powered by your own LLM.

⚡️ Quick Start

1. Install

npm install -g alif-digest

2. Initialize

Alif will interactively guide you through connecting an LLM (Local Ollama, Anthropic, or OpenRouter) and setting up your delivery channels.

alif init

3. Run

Alif scrapes all sources, applies multi-layer scoring, analyzes the top items with your LLM, and delivers the results.

alif run

# Skip source cooldown (re-fetch all sources immediately)
alif run --force

4. Schedule

Keep the signals flowing automatically.

alif schedule add     # Schedule a recurring digest
alif schedule list    # View scheduled jobs
alif schedule check   # Trigger any pending scheduled jobs

✨ Features

• Multi-Layer AI Scoring: Specialized two-stage evaluation engine combining keyword heuristics and deep NLP analysis to surface truly important signals.
• Smart Deduplication: Blazing-fast fuzzy clustering ensures you never read three variations of the same news story.
• Aggressive Noise Reduction: Automatically downranks PR pieces, sponsored fluff, and empty listicles.
• Broad Content Ingestion: Seamlessly scrapes various RSS, JSON API, raw HTML, and ArXiv feeds.
• Customizable Delivery: Push your personalized daily digests directly to Slack or any Webhook.
• Bring Your Own AI: Works out-of-the-box with local Ollama models or cloud providers like OpenRouter.
• Built-in Automation: Schedule recurring digests to keep your feeds curated entirely on autopilot.

🤖 Models

Use standard instruction-tuned models that support structured output. Avoid thinking/reasoning models (e.g. DeepSeek R1, Qwen Thinking variants) as they interfere with JSON schema generation.

Tested Models

| Provider | Model | | -------------- | -------------------- | | Ollama | llama3.2:3b | | OpenRouter | openai/gpt-4o-mini |

🛠 CLI Reference

alif init

Interactive wizard to configure your LLM provider, model, delivery channel, and preferences.

alif validate

Runs a strict pre-flight check on your config.json and feeds.json files. It verifies that your Cloud LLM API keys are present (if selected), that Delivery Webhook URLs are valid HTTP destinations, and that the feeds.json array exists and contains the correct format schema.

alif run [--force] [--verbose] [--quiet]

Run the full pipeline.

• --force: Bypass the source cooldown to re-fetch all sources immediately.
• --verbose: Stream detailed layer-by-layer scoring and prompt output (great for debugging LLMs).
• --quiet: Suppress all output except for errors.

alif config

Manage your configuration without manually editing config.json.

alif config show                          # Print current config
alif config set signalThreshold 70        # Update a preference value
alif config set maxItemsPerRun 5          # Change how many items are delivered
alif config set sequentialAnalysis true   # Enable one-by-one LLM processing
alif config toggle-ai-scoring             # Toggle Layer 2 AI Article Scoring on/off
alif config set logLevel verbose          # Change log verbosity (silent, normal, verbose)
alif config set noColor true              # Disable ANSI colored output

alif schedule

alif schedule add     # Add a new cron schedule
alif schedule list    # List all schedules
alif schedule delete  # Remove a schedule
alif schedule check   # Run any due schedules

alif debug

alif debug llm <provider> --model <name> [--key <api-key>]
# Test your LLM connection and see the full audit trail (prompt, response, latency)

alif debug audit-feeds [--output <path>]
# Scrape all feeds dry-run: see item counts and content sizes per source
# Saves a detailed JSON report (default: ~/.config/alif/audit_report.json)

⚙️ Configuration Reference

Config is stored at ~/.config/alif/config.json. Most values can be changed with alif config set.

| Preference | Type | Default | Description | | ------------------------- | --------- | ---------- | ---------------------------------------------------------------- | | signalThreshold | number | 60 | Minimum score (0–100) for an article to qualify | | maxItemsPerRun | number | 10 | Max articles delivered per run | | sourceCooldownMinutes | number | 5 | Minimum gap between re-fetching the same source | | sequentialAnalysis | boolean | false | Analyze articles one-by-one (recommended for small local models) | | enableAIArticlesScoring | boolean | true | Enable Layer 2 LLM-based article scoring | | customKeywords | object | {} | Add or override keyword signal weights | | negativeKeywords | object | {} | Add custom noise/penalty keywords | | logLevel | string | 'normal' | Set logging verbosity: 'silent', 'normal', or 'verbose' | | noColor | boolean | false | Disable colored terminal output (NO_COLOR=1 also works) |

Custom keywords example:

"customKeywords": {
  "my-framework": 30,
  "competitor-name": 0
},
"negativeKeywords": {
  "webinar": 20
}

☁️ Running on GitHub Actions (Free)

To ensure your API keys and workflow logs remain completely secure and hidden from the public, we highly recommend running this in a Private GitHub repository.

1. Create a new Private Repository on GitHub.
2. Copy the example workflow: Copy the examples/alif-scheduled-run.yml file into your new private repository at .github/workflows/alif-scheduled-run.yml.
3. Navigate to your private repository's Settings > Secrets and variables > Actions.
4. Add your sensitive access credentials as Repository Secrets.
5. Configure your preferences as Repository Variables (optional).

Environment Variables Reference

| Variable | Type | Possible Values | Default | Description | | ------------------------------ | -------- | ----------------------------------------- | ------------------- | ----------------------------------------------------------------------------- | | ALIF_LLM_PROVIDER | Variable | 'anthropic', 'openrouter', 'ollama' | 'anthropic' | The AI provider to use for analysis. | | ALIF_LLM_API_KEY | Secret | Your API Key | - | Universal key. Alternatively use ANTHROPIC_API_KEY or OPENROUTER_API_KEY. | | ALIF_LLM_MODEL | Variable | e.g. 'claude-3-5-sonnet-20240620' | Provider Specific | The specific model to use. | | SLACK_WEBHOOK_URL | Secret | https://hooks.slack.com/... | - | Webhook URL for Slack delivery. | | GENERIC_WEBHOOK_URL | Secret | https://... | - | Webhook URL for generic server delivery. | | ALIF_SIGNAL_THRESHOLD | Variable | 0 - 100 | 60 | Minimum score for an article to qualify. | | ALIF_MAX_ITEMS_PER_RUN | Variable | 1+ | 10 | Max articles delivered per run. | | ALIF_SOURCE_COOLDOWN_MINUTES | Variable | 0+ | 5 | Minimum gap between re-fetching a source. |

The included workflow automatically runs every day at 8:00 AM UTC (you can change the schedule in the file), caches your database state securely, and delivers your customized digest!

🤝 Architecture & Contributing

Interested in how Alif works under the hood or want to run it locally?

Check out our Contributing Guidelines for information on the architecture, local development setup, and how to submit pull requests!

📢 Acknowledgements

Thanks to Roland for the initial technical foundation, and Antigravity for the help in building this project.

📄 License

Distributed under the MIT License. See LICENSE for more information.