Navier-Stocks is a high-performance, heterogeneous multi-agent algorithmic trading system engineered specifically for the Indian Stock Exchanges (NSE & BSE).
Built from the ground up in Go, the system treats market liquidity, order book momentum, and price velocity as structural fluid vectors. By treating volatile market movements as a bounded computational problem governed by predictable mechanical thresholds—much like navigating the immersed boundaries of a complex fluid simulation—Navier-Stocks isolates high-probability intraday momentum breakouts while strictly managing capital correlation.
Designed for extreme operational efficiency, the entire distributed architecture is optimized to run flawlessly on constrained, hobby-tier cloud infrastructure with sub-5 millisecond internal routing latency.
Navier-Stocks achieves its speed and resilience by strictly decoupling the execution pipeline into two distinct streams: a Fast Path for deterministic mathematical execution, and a Slow Path for asynchronous cognitive context.
Agents communicate via NATS JetStream for low-latency, fault-tolerant message passing, while PostgreSQL anchors the system's long-term state and operational memory.
| Agent | Paradigm | Role & Responsibility |
|---|---|---|
| 📡 Quantitative Scout | Deterministic / Go | The real-time ingestor. Subscribes to live WebSocket ticks, maintaining hyper-efficient in-memory ring buffers. It solves for pressure differentials in the market, hunting for VWAP + Bollinger Band momentum breakouts. |
| 🧠 Sentiment Analyst | Cognitive / LLM | The asynchronous macro observer. It continuously scrapes and processes financial news and corporate filings, writing contextual confidence scores to the database without ever blocking the fast execution path. |
| 🛡️ Risk Manager | Deterministic / Go | The structural gatekeeper. It intercepts trade intents, validating them against rigid intraday time windows, lock-free sector concentration maps, and overall portfolio limits before calculating precise capital allocation. |
| ⚡ Trader Gateway | Interface | The "dumb pipe" execution layer. Seamlessly hot-swappable between a PaperTrader for database-only simulation and a GrowwTrader for live REST API broker interactions. |
Standard multi-agent systems bottleneck when forcing LLMs to make real-time decisions. Navier-Stocks sidesteps this by keeping the core execution loop (Scout -> Risk Manager -> Trader) strictly mathematical. By leveraging Go's lightweight goroutines and NATS byte-payloads, a live market tick is ingested, validated, and translated into an execution signal in under 5 milliseconds.
Intraday diversity is defined by correlation control. The Risk Manager utilizes a sync.RWMutex state map to instantly enforce:
- Sector Concentration Caps: Preventing exposure to systemic shocks by limiting active trades per structural sector (e.g., maximum one position in Nifty IT or Nifty Bank at a time).
- Concurrency Ceilings: Capping the total number of simultaneous open positions to preserve capital liquidity.
- Directional Equilibrium: Ensuring the ratio of Long vs. Short positions remains balanced against broader index momentum.
Capital preservation is hardcoded into the platform's DNA. Every execution boundary is calculated dynamically using a fixed-fractional allocation matrix. The system ensures that a technical stop-loss trigger will never jeopardize more than a predefined fraction of the total deployable capital:
Built for the specific mechanics of the Indian markets, the system programmatically sidesteps opening volatility and broker-enforced liquidation penalties:
- Opening Buffer: Execution suppression from 09:15 to 09:30 AM IST to allow the boundary layer of the market to stabilize.
- Closing Cutoff & Auto-Square-Off: Halts new entries post-15:00 IST and automatically triggers market-liquidation for all open MIS positions precisely at 15:15 PM IST.
- Language: Go (Golang) — Chosen for unparalleled concurrency (
goroutines), minimal memory footprint, and rapid binary execution. - Message Broker: NATS JetStream — Provides persistent, at-least-once delivery pub/sub messaging for agent communication, outperforming Redis in low-memory environments.
- Database: PostgreSQL — Handles ACID-compliant state recovery, historical transaction auditing, and asynchronous sentiment parameter storage.
- Orchestration: Docker & Docker Compose — Containerized deployments ensuring exact parity between local development and cloud production environments.
- Docker and Docker Compose
- Go 1.27+ (For local development outside of containers)
- Clone the repository:
git clone https://github.com/Nalin-Angrish/Navier-Stocks.git
cd navier-stocks- Configure the environment:
cp .env.example .env
# Edit .env with your broker API keys, database credentials, and risk parameters- Deploy the infrastructure stack:
docker-compose up -d --buildThis single command provisions the Go agent microservices, initializes the PostgreSQL schemas, and spins up the NATS JetStream broker.
- Verify System Health:
docker-compose logs -f risk-managerFor a deep dive into the underlying quantitative math, NATS JSON payloads, or operational runbooks, please consult the official Navier-Stocks Engineering Wiki.
Navier-Stocks is strictly an educational engineering project and a Proof-of-Concept for distributed multi-agent systems. This repository does not constitute financial advice. Algorithmic trading on live financial markets carries a high risk of catastrophic capital loss. The creators of this repository are not responsible for any financial losses incurred through the deployment of this software. Always test thoroughly using the PaperTrader sandbox interface before supplying live broker credentials.