Your personal intelligence desk.
Monitors the news. Builds the knowledge graph. Writes the briefing. Records the podcast.
Every morning, before you wake up.
GitHub · Docs · Live Demo · System Map · Quick Start
Nexus monitors news sources across languages, extracts structured events, builds a persistent knowledge graph, and turns it all into daily briefings, podcast episodes, and probabilistic forecasts. Delivered to your phone or explored through a live dashboard.
You pick the topics. Nexus reads everything, decides what matters, connects the dots, and tells you what changed.
Daily briefing + podcast. A concise intelligence report synthesized from dozens of sources, with a two-host podcast episode generated via TTS. Delivered to Telegram every morning at a time you choose.
Live dashboard. Explore topics, narrative threads, entities, events, convergence/divergence signals, source balance, cost tracking, and filter transparency. Everything Nexus knows is browsable.
Knowledge graph. Entity resolution across sources and languages. Persistent narrative threads that track how stories evolve over days and weeks. Convergence detection when independent sources confirm the same fact; divergence when they frame it differently.
Forward Look (Beta). Actor-based probabilistic forecasts generated from your knowledge graph. Optional Kalshi market comparison when prediction market data is available.
Breaking news. Between daily runs, Nexus polls wire feeds and alerts you when something crosses your significance threshold.
Ask it questions. Text the Telegram bot a question about any tracked topic. It answers from accumulated knowledge, falling back to web search when needed.
git clone https://github.com/zantryis/NEXUS.git
cd NEXUS
docker compose upRequires Python 3.11+ and ffmpeg (for podcast audio).
git clone https://github.com/zantryis/NEXUS.git
cd NEXUS
pip install -e ".[all]"
python -m nexus runOpen http://localhost:8080 and the setup wizard walks you through everything: provider, API key, topics, timezone, schedule, and optional Telegram. Source discovery, the first pipeline run, historical backfill, and Forward Look generation all happen automatically.
One LLM API key. The wizard helps you choose:
| Preset | Provider | ~Cost/Day |
|---|---|---|
free |
Ollama (local) | $0 |
cheap |
DeepSeek | $0.01 |
balanced |
Gemini Flash + Pro | $0.05 |
quality |
Gemini Pro | $0.15 |
openai-cheap |
GPT-5.4 Mini | $0.05 |
openai-balanced |
GPT-5.4 Mini + 5.4 | $0.15 |
openai-quality |
GPT-5.4 Mini + 5.4 | $0.30 |
anthropic |
Claude Sonnet 4.6 | $0.15 |
Optional: TELEGRAM_BOT_TOKEN for phone delivery (@BotFather) · ELEVENLABS_API_KEY for alternative TTS
20 pipeline stages take raw RSS all the way to briefings, podcasts, and forecasts. Everything persists to a single SQLite file. No external databases, no cloud infrastructure beyond the LLM API. Back up the system by copying one folder.
No API keys needed. Explore the dashboard with sample data:
python -m nexus demo seed --from-scratch
python -m nexus demo serveDemo mode is read-only: settings locked, all pages render with seeded data. Good for screenshots and walkthroughs.
| Key | When Needed |
|---|---|
GEMINI_API_KEY |
Gemini presets |
OPENAI_API_KEY |
OpenAI presets |
ANTHROPIC_API_KEY |
Anthropic preset |
DEEPSEEK_API_KEY |
DeepSeek preset |
TELEGRAM_BOT_TOKEN |
Telegram delivery |
ELEVENLABS_API_KEY |
ElevenLabs TTS |
NEXUS_ADMIN_TOKEN |
Remote access to /setup and /settings |
At least one LLM key is required (unless using the free/Ollama preset).
Four topics ship with pre-built source registries: Iran-US Relations, AI/ML Research, Formula 1, and Global Energy Transition. For custom topics, Nexus auto-discovers RSS feeds during setup.
Each topic supports subtopics, source languages, filter thresholds (1-10), scope (narrow/medium/broad), and perspective diversity settings. See data/config.example.yaml for the full schema.
- Binds to
127.0.0.1by default, only reachable from your machine - Docker Compose adds a narrow gateway exception for
localhost:8080 - For remote access: set
NEXUS_ADMIN_TOKENand openhttp://host:8080/settings?admin_token=YOUR_TOKENonce - Setup wizard is disabled after first run unless
NEXUS_ALLOW_SETUP_RESET=1
Cap ingestion for fast testing:
NEXUS_SMOKE_MODE=20 python -m nexus run # 20 articles per topic maxpython -m nexus run # Start all services (dashboard + scheduler + bot)
python -m nexus engine # Run pipeline once
python -m nexus serve # Dashboard only
python -m nexus setup # Interactive CLI setup wizard
python -m nexus sources discover <slug> # Auto-discover RSS feeds for a topic
python -m nexus sources check # Test RSS feed health
python -m nexus forecast generate # Generate Forward Look forecasts
python -m nexus demo seed [--from-scratch] # Seed demo database
python -m nexus demo serve # Start demo serverRun python -m nexus --help for the full command list including benchmark, evaluation, and experiment tooling.
The prediction surface generates calibrated probabilistic forecasts from your knowledge graph using an actor-based reasoning engine. For each question, Nexus identifies relevant actors, reasons about their likely behavior, and synthesizes a probability estimate. Raw outputs are compressed toward 0.5 to correct for LLM overconfidence (gamma=0.8).
When Kalshi prediction market data is available, Nexus generates independent estimates for the same questions and shows the comparison. Useful for spotting where your knowledge graph sees something the market doesn't.
The bigger forecast lab (6 engines, hindcast framework, benchmark suites) ships in-repo but is not part of the default setup or onboarding. See ARCHITECTURE.md for details.
pip install -e ".[all,dev]"
pytest -m "not e2e and not integration" # Default test suite (1,548 unit tests)
pytest -m e2e # E2E tests (99 tests, requires API keys)
python -m nexus experiment # Internal benchmark suitesSee ARCHITECTURE.md for the full module map, schema details, and contributor reference.
Built by @zantryis