Open Source Agent Architecture
Automate Equity Research without the Cloud API Tax.
Piping raw news articles and financial filings directly into cloud LLMs is cost-prohibitive at scale. StoxFlow implements a hybrid multi-agent architecture using LangGraph, compressing text locally with Ollama before synthesizing the final thesis via Gemini.
View the Repository
account_tree LangGraph Pipeline
- check Resolves semantic ticker queries
- check Fetches Upstox API fundamentals concurrently
- check Processes technical weekly OHLC candles
- check Extracts signals via local Qwen SLMs
- check Synthesizes thesis via Cloud LLM
- check Observability telemetry via Arize Phoenix
savings Cost Optimization
Instead of sending 25,000+ tokens of raw news directly to a frontier model, StoxFlow compresses the articles locally on your machine. This reduces cloud API input token consumption by over 80%.
// Hybrid Execution Flow:
1. Scrape Web DOM (Async HTTP)
2. Extract Facts & Sentiment (Ollama 3B)
3. Synthesize JSON Report (Gemini Flash)