17sec — Autonomous News Intelligence Platform

17 seconds. That's the delay between reality and cable news. We're building what comes before.

What It Does

An autonomous agentic news platform that:

1. Detects events before they're "news" (anomaly detection on signal streams)
2. Spins up agents to gather context, verify, and build narrative
3. Publishes situations with headlines, context, forecasts
4. Exposes API for other dashboards, newsrooms, trading systems
5. Learns from resolution — calibrated forecasts with receipts

Tech Stack

• Backend: Python 3.11+, FastAPI, SQLAlchemy (async), PostgreSQL + pgvector
• Queue: Redis Streams for real-time event distribution
• Real-time: Server-Sent Events (SSE)
• Search: Meilisearch
• LLM: Anthropic Claude (Haiku for speed, Sonnet for depth)
• Frontend: Next.js 14, Tailwind CSS, Zustand, Recharts
• Infra: Docker Compose

Quick Start

# 1. Clone and configure
cp .env.example .env
# Add your ANTHROPIC_API_KEY

# 2. Start infrastructure
docker-compose up -d postgres redis meilisearch

# 3. Backend
cd backend
pip install -e .
python -m spacy download en_core_web_sm
alembic upgrade head
python -m scripts.seed

# 4. Run backend
uvicorn app.main:app --reload

# 5. Start worker (new terminal)
python -m app.workers.main

# 6. Frontend (new terminal)
cd ../frontend
npm install
npm run dev

Architecture

Signal Sources (RSS, X API, News APIs, GDELT)
    ↓
Ingest Workers (dedup, enrich, score themes)
    ↓
Detection (anomaly detection, clustering)
    ↓
Situations (grouped events with status)
    ↓
Agent Flows (headline, summary, context, forecast)
    ↓
API + SSE → Frontend / External Dashboards

Agent System

12 autonomous agents organized in 3 tiers:

• Detectors: anomaly_detector, cluster_detector, verified_source_detector
• Investigators: geo_investigator, entity_investigator, timeline_builder, verification_agent, context_agent
• Publishers: headline_agent, summary_agent, forecast_agent, juxtaposition_agent

Agents are configurable via admin panel — model, temperature, max tokens, and prompt templates are all editable without code changes.

API

• GET /api/situations — Active situations
• GET /api/situations/breaking — Breaking/crisis situations
• GET /api/forecasts — Open forecasts
• GET /api/forecasts/calibration — Calibration data (the moat)
• GET /api/themes — Theme temperatures
• GET /api/stream — SSE real-time feed
• GET /api/embeds/ticker — Embeddable ticker
• Admin endpoints for agent management, prompt editing, run history

What Makes This Different

1. Agents — not just RSS aggregation, autonomous investigation
2. Situations — clustered events, not firehose
3. Forecasts — calibrated predictions with track record
4. Editable Prompts — admin can tune without code
5. API-first — infrastructure for other dashboards
6. Verification — clear confidence levels
7. Transparency — agent reasoning visible