LynxInvest
An agentic investment engine that turns news events into signals and produces 24/7 investment decisions through a multi-agent decision pipeline (paper + Binance testnet).
Overview
Scores the news flow with an LLM and turns it into automated, reasoned, and traceable investment decisions through a disciplined ensemble of agents and risk guardrails. The goal is not a "tool that finds the best model" but an investment tool that is measured and learns in a closed loop.
Not a "who predicts better" contest, but the reasoned consensus of four disciplines + absolute guardrails; proactive scanning that produces even without news; the idea of tying the model arena to live capital allocation.
Working / production-like; validated live end-to-end — diversified trades in paper mode + real (fake-money) orders on Binance testnet. Comprehensive test suite.
Highlights
Momentum Hunter, Value Investor, Risk Keeper, Contrarian — look at the same signal through four different investment disciplines; an aggregator makes the final call with absolute guardrails.
Crypto (Binance) + equities/FX/commodities/indices (Yahoo Finance) pricing; no longer BTC-only, it diversifies.
Periodically scans the structured asset universe even when no news arrives — a non-news technical signal path.
Goal-driven progress tracking (e.g. $100 → $110 / 1 week) + mandate-aware position sizing.
Drawdown protection, stop-loss, position weight cap, spot-only, cash checks, Risk Keeper veto.
A risk-adjusted P&L table where different LLM providers race toward the same financial goal.
Each decision is measured against the realized outcome over 1h/24h/72h horizons.
NAV chart, mandate progress card, P&L and positions.
Tech Stack
| Backend | Python + FastAPI |
| Task queue | Celery + beat |
| Data | PostgreSQL · Redis |
| AI | Multi-provider via LiteLLM (local/free + cloud) |
| Pricing | Binance & Yahoo public API (keyless) |
| Infrastructure | 24/7 self-host with Docker |