██████╗  █████╗ ██████╗  █████╗ ██╗     ██╗      █████╗ ██╗  ██╗
  ██╔══██╗██╔══██╗██╔══██╗██╔══██╗██║     ██║     ██╔══██╗╚██╗██╔╝
  ██████╔╝███████║██████╔╝███████║██║     ██║     ███████║ ╚███╔╝
  ██╔═══╝ ██╔══██║██╔══██╗██╔══██║██║     ██║     ██╔══██║ ██╔██╗
  ██║     ██║  ██║██║  ██╗██║  ██║███████╗███████╗██║  ██║██╔╝ ██╗
  ╚═╝     ╚═╝  ╚═╝╚═╝  ╚═╝╚═╝  ╚═╝╚══════╝╚══════╝╚═╝  ╚═╝╚═╝  ╚═╝
Open Source

Personal AI research agent.
Collects signals from 8 web sources, scores them with LLM,
combines news + market data into actionable briefings — from CLI or Telegram.

Get Started View on GitHub
8Web Sources
1–10CCW AI Scoring
RAGKnowledge Base
CLI + TGInterfaces

Everything you need to stay ahead

One agent that collects, scores, and synthesizes — so you don't have to.

Web Intelligence

Collects from Hacker News, Reddit, GitHub Trending, Product Hunt, Mastodon, Dev.to, Google Trends, and custom RSS in parallel. Deduplicates via MD5.

RAG Knowledge Base

Index local files (PDF, DOCX, CSV, MD) into ChromaDB. Ask questions — answers come from your own documents combined with collected intel.

CCW Impact Scoring

Every item is scored 1–10 by LLM for real-world impact. Scores rank news, prioritize briefing context, and trigger smart alerts at 7+.

Market Data

Live crypto (CoinGecko), stocks (Alpha Vantage + Finnhub), and forex (open.er-api). All free APIs. SQLite history for trend tracking. Auto-alerts on big moves.

Daily Briefing

Structured report: signal themes, market analysis, dominant trend, capital flow, BUY/HOLD/AVOID verdict, and 3 specific things to research today.

Telegram Bot

Full command set in Telegram. Send a file → indexed. Smart alerts pushed automatically. Morning briefing at your configured hour. User whitelist via TELEGRAM_ALLOWED_USER_ID.

CCW Impact Score

Every collected item is scored by the LLM before it reaches you.

9–10
Major breakthrough / civilisation-level event
7–8
Significant shift worth acting on
5–6
Interesting, worth following
1–4
Routine noise

Get started in minutes

Works with local Ollama or Anthropic API — your choice.

1

Clone & install

git clone https://github.com/ysz7/Parallax.git
cd Parallax
python -m venv .venv
.venv\Scripts\activate        # Windows
source .venv/bin/activate     # Linux / macOS
pip install -r requirements.txt
2

Pull models (Ollama) or set API key

ollama pull gemma3
ollama pull nomic-embed-text

Or use Anthropic — set LLM_PROVIDER=anthropic and ANTHROPIC_API_KEY in .env. Recommended for server deployments (no GPU needed).

3

Configure .env

LLM_PROVIDER=ollama
OLLAMA_MODEL=gemma3

# For Telegram bot:
TELEGRAM_BOT_TOKEN=your_token
TELEGRAM_ALLOWED_USER_ID=123456789

# Auto-research every 6 hours:
AUTO_RESEARCH_INTERVAL=360
4

Run

python cli.py   # local CLI
python bot.py   # Telegram bot

Windows shortcuts: start-cli.bat and start-bot.bat auto-activate venv.

Commands

/researchCollect all sources + CCW scoring + market snapshot
/news [N]Last N items sorted by CCW score
/briefing [days]Full briefing: signals + market + investment direction
/predict [days]Cross-niche predictions with probabilities
/flows [days]Sectors where money is moving now
/ideas5 business ideas based on current trends
/marketCrypto + stocks + forex live snapshot
/report <topic>Deep research report on any topic
/remind <text>Natural language reminders (en + ru)
/statsKnowledge base stats + manage indexed files
/tokensAnthropic API usage and cost
/set lang <code>Response language: EN, RU, DE, etc.

Any text → RAG search. File path or drag & drop → indexed automatically.

Deploy with Docker

Run the Telegram bot on any VPS — no GPU needed when using Anthropic.

Data persists in ./data/ on the host. Just update .env and restart.

docker compose up -d
# docker-compose.yml
services:
  bot:
    build: .
    env_file: .env
    volumes:
      - ./data:/app/data
    restart: unless-stopped