OpenVC-AI

OpenVC-AI is an A2A-style multi-agent investment research platform. This MVP focuses on public-market forecasting first, then can be expanded into a startup/venture-capital research platform.

What it does now

• Uses Groq first for LLM analysis.
• Can fall back to OpenAI if configured.
• Pulls historical stock prices from Alpha Vantage.
• Pulls recent company news from Finnhub.
• Runs a baseline Monte Carlo forecast using historical log returns.
• Uses a News Analysis Agent to summarize bullish, bearish, and unknown factors.
• Exposes a FastAPI API.
• Keeps secrets out of code with .env.

Architecture

User / API Client
  |
FastAPI Controller
  |
Central A2A Orchestrator
  |
  |-- QuantForecastAgent
  |-- NewsAnalysisAgent
  |-- InvestmentMemoAgent
  |
Ports / Adapters
  |-- GroqLLMAdapter
  |-- OpenAILLMAdapter
  |-- AlphaVantageMarketDataAdapter
  |-- FinnhubNewsAdapter

This is not a full external A2A network yet. The local agents are structured so they can later be converted into remote A2A agents with agent cards and discovery.

Setup

cd openvc-ai
python3 -m venv .venv
source .venv/bin/activate
pip install -e .
cp .env.example .env

Edit .env and add your new rotated keys. Do not use any key that was pasted into chat.

Run

python -m openvc_ai.main

Open:

http://localhost:8080/docs

Example request

curl -X POST http://localhost:8080/forecast \
  -H "Content-Type: application/json" \
  -d '{"ticker":"NVDA","horizon_days":30,"include_news":true}'

Endpoints

Health

GET /health

Agent Cards

GET /agents

Forecast

POST /forecast

Body:

{
  "ticker": "NVDA",
  "horizon_days": 30,
  "include_news": true
}

Next features to add

1. Real A2A agent server/client transport.
2. MCP tool server for market data and news tools.
3. Startup discovery agent.
4. Founder analysis agent.
5. Competitive landscape agent.
6. SEC filings adapter.
7. FRED macroeconomic adapter.
8. Postgres + pgvector memory.
9. Backtesting engine.
10. GitHub automation agent for open-source repo generation.

Safety note

This project produces research-style outputs only. It is not financial advice and should not be used as an automated trading system without validation, risk controls, and compliance review.