TradingAgents (TauricResearch/TradingAgents) is an open-source Python framework that models a trading firm as a multi-agent system. Specialist agents — fundamental analyst, sentiment analyst, news analyst, technical analyst, bull/bear researchers, trader, and a risk-management committee — communicate through a structured debate-and-synthesis loop, then hand decisions to a portfolio manager. Built on LangGraph, it supports Claude, GPT-4, and DeepSeek as underlying models, and ships with backtesting, paper-trading, and pluggable data adapters (Yahoo Finance, Alpha Vantage, custom APIs). Trended +3,315 stars on May 4, 2026 — one of the fastest-rising vertical agent frameworks in the May 2026 directory wave, with 64,746 total stars. Research-first: the authors explicitly position it as a study of multi-agent coordination in financial domains, not a live trading product.
TradingAgents models a trading firm as a multi-agent system, with specialist agents — fundamental analyst, sentiment analyst, news analyst, technical analyst, bull and bear researchers, trader, and a risk-management committee — passing structured signals and debate transcripts through a LangGraph orchestration. Where most multi-agent demos collapse into single-LLM-with-multiple-prompts, TradingAgents preserves the adversarial structure: the bull and bear researchers actually argue with each other before the trader makes a call, and the risk committee can override.
The repo crossed 64,746 stars on May 4, 2026, with +3,315 added in the last 24 hours — the steepest climb in the vertical-agent wave that’s reshaping the AgentConn directory this week. The framework’s appeal is twofold: it’s a working research artifact for studying multi-agent coordination, and the financial domain is constrained enough that you can verify whether the agents produced something coherent.
Three-layer agent topology:
Each agent emits structured outputs that downstream agents can parse — no free-form prose-to-prose chains. The LangGraph backbone makes the orchestration auditable: every decision has a complete signal-to-debate-to-trade chain.
Multi-agent research: The cleanest reference architecture currently available for adversarial multi-agent reasoning. Researchers cite it as a baseline for debate-style agent coordination.
Financial AI prototyping: Teams building proprietary trading research can swap in custom data adapters and proprietary signal sources while keeping the orchestration scaffolding.
Backtesting and education: The codebase ships clean backtests against historical data — a useful teaching tool for understanding how multi-agent decisions compose over time.
This is a research framework, not a live trading product. There’s no execution layer, no compliance/regulatory infrastructure, and no real-time data feeds out of the box. The authors are explicit: TradingAgents demonstrates how multi-agent systems can analyze and reason about financial data, not how to automate live trading. Use it for research, education, and prototyping — not for production capital allocation.
Open-source multi-agent framework for AI-driven financial analysis — simulates a hedge fund with specialized analyst agents and a portfolio manager agent.
Incremental data framework for AI agents — ingestion and indexing with Rust-grade performance and propagation of changes across joins.
TypeScript deep financial research agent that produces investor-grade equity research reports — autonomously gathering data, modeling, and writing analyst-quality output.