Agentic Trading Lab is an open-source experimental platform for LLM-powered trading agents.
Turn trading ideas into traceable experiments: prototype agents, run backtests and paper-trading simulations, inspect reasoning and decision logs, benchmark against market baselines, and study how agents behave under near-real financial constraints.

Outline

• Overview
• Key Features
• Run Locally
• File Structure
• User Workflow
• Roadmap
• Citation
• License
• Contributing

Overview

Agentic Trading Lab is an interactive research and educational platform for exploring trading systems powered by large language models. Developed alongside a systematic survey of agentic trading research, it helps students, researchers, and developers explore how agents reason, trade, and perform in realistic market environments. Move beyond backtest returns by customizing agents, inspecting their decisions, evaluating risk, progressing from historical simulation to live-market paper trading, and comparing performance on standardized leaderboards.

Key Features

Talk to agents. Test trading ideas. Start with a market question or trading idea, then turn it into an agent you can run, observe, and improve.

• Create trading agents your way
  Choose a model, data source, and trading prompt, or connect your own agent through our API.
• Start from a template
  Add a ready-made agent from the Community tab, then customize its prompts and backtest it.
• Build an agent-powered portfolio
  Give multiple agents simulated capital and manage them as your own virtual trading team.
• Test before using real capital
  Move from historical backtests to live-market paper trading in one workflow.
• Go live when you're ready
  Connect a real Robinhood account and let an agent trade it under per-order risk caps. Off by default, with a review-only mode that shows what an agent would do before it can act — see Live Trading.
• See every run, decision, and reason
  Monitor positions, trades, portfolio changes, and the reasoning behind each action.
• Measure more than returns
  Evaluate performance, risk, drawdown, and trading behavior using standardized metrics.
• Compare agents in the open
  Benchmark LLM models, baseline strategies, and market indices on an open leaderboard under the same market window.

Run Locally

uv venv --python 3.13 .venv
UV_HTTP_TIMEOUT=180 uv pip install --python .venv/bin/python -r requirements.txt -r requirements-vnpy.txt pytest pytest-timeout
dashboard/scripts/run_local_safe.sh        # http://127.0.0.1:8765/app, no keys, nothing can trade or bill

Full guide (backtests, tests, credentials, troubleshooting): docs/local-run.md.

File Structure

AgenticTrading/
├── dashboard/                 # Shipping product
│   ├── backend/               # FastAPI package (dashboard.backend.*)
│   │   ├── api/               # /api routers + Agent API v2 (/api/v2)
│   │   ├── domain/            # Business logic (runs, backtesting, leaderboard, trading, …)
│   │   ├── execution/         # v2 execution backends (backtest live; paper stub)
│   │   ├── infrastructure/    # LLM validator, market data, Alpaca + Robinhood brokers
│   │   └── integrations/      # Discord bot, etc.
│   ├── frontend/              # Static assets: landing (/) + dashboard (/app)
│   ├── landing/               # Vite/React landing source (builds into frontend/)
│   ├── scripts/               # CLI backtests (backtest_hourly_agent.py, …)
│   ├── config/                # defaults.json, leaderboard.json, marketplace.json
│   └── storage/               # data/backtest.db + backups/
├── packaging/agentictrading/  # PyPI SDK (AgentRunner + HTTP client)
├── credentials/               # Local only — not in git (see alpaca.json.example)
├── docs/                      # Sphinx docs + architecture notes
├── orchestration/             # FinAgent multi-agent framework (separate)
├── requirements.txt           # Dashboard deps (not root pyproject.toml)
├── Dockerfile / render.yaml   # Backend deploy
└── dashboard/frontend/vercel.json  # Frontend static deploy (must sit in the Vercel Root Directory)

User Workflow

Roadmap

Agentic Trading Lab is evolving from an experimental playground into an open platform where trading agents, models, data, and tools can be connected, managed, evaluated, and reused.

• Standardized agent integration
  Define reusable Agent Cards, manifests, runners, trace schemas, and evaluator hooks for connecting agents to the Lab.
• Broader agent and tool connectivity
  Connect popular open-source trading projects, model providers, data providers, brokers, and external agents through adapters, SDKs, and MCP.
• Managed agent runtime
  Support scheduled and long-running agents with persistent state, monitoring, alerts, and failure recovery.
• Dynamic market evaluation
  Evaluate agents through live-market paper trading with execution records, costs, risks, and reasoning traces—not only final returns.
• Risk and stress testing
  Test agents under changing market conditions with configurable risk policies, approval rules, and kill switches.
• Open agent ecosystem
  Grow Agent Cards, reproducible submissions, competitions, and the open leaderboard toward hundreds and eventually thousands of reusable agent instances.

Progress so far: news sentiment is live via Agentic FinSearch (Home panel + v2 agent context), with Reddit/social sentiment still planned. Under Broader agent and tool connectivity, the first external-framework adapter has shipped — TradingAgents runs locally on your own model and data credentials and replays its decisions through the Python SDK (setup guide, written in Simplified Chinese) — while MCP, broker, and data-provider adapters are still ahead.

Citation

This repository includes the FinAgent Orchestration Framework under orchestration/, developed by Jifeng Li et al. at Open Finance Lab. The orchestration framework provides multi-agent architecture, memory systems, and DAG-based planning components. See orchestration/README.md for details.

If you use the orchestration framework in research, please cite:

@inproceedings{orchestration_finagents_2025,
   title     = {Orchestration Framework for Financial Agents: From Algorithmic Trading to Agentic Trading},
   author    = {Jifeng Li and Arnav Grover and Abraham Alpuerto and Yupeng Cao and Xiao-Yang Liu},
   booktitle = {NeurIPS 2025 Workshop on Generative AI in Finance},
   year      = {2025},
}

Plain-text citation:

Jifeng Li, Arnav Grover, Abraham Alpuerto, Yupeng Cao, and Xiao-Yang Liu. Orchestration Framework for Financial Agents: From Algorithmic Trading to Agentic Trading. NeurIPS 2025 Workshop on Generative AI in Finance, 2025.

Documentation: finagent-orchestration.readthedocs.io. Local preview: docs/README.md

License

OpenMDW-1.0 — See LICENSE (Copyright @ SecureFinAI Lab)

Contributing

Pull requests and issues welcome!

---

Built with Alpaca API, FastAPI, Chart.js, and SQLite