🚀 AI Trading - PyTrade's Documentation

AI-Trading Apps¶

Disclaimer: I am the creator of 🐘 alphafund

REPOS = {
    "alphafund": {
        "github": "PFund-Software-Ltd/alphafund",
        "_fetch_github": False,
        "Status": "🚧",
        "Approach": "💻 Code-to-Trade",
        "Backtest": True,
        "Live": True,
        "Research": True,
        "TradFi": True,
        "CeFi": True,
        "DeFi": False,
        "Prediction Market": False,
    },
    "nofx": {
        "github": "NoFxAiOS/nofx",
        "Status": ProjectStatus.ACTIVE,
        "Approach": "Prompt-to-Trade",
        "Backtest": True,
        "Live": True,
        "Research": True,
        "TradFi": False,
        "CeFi": True,
        "DeFi": False,
        "Prediction Market": False,
    },
    "vibe-trading": {
        "github": "HKUDS/Vibe-Trading",
        "Status": ProjectStatus.ACTIVE,
        "Approach": "Prompt-to-Trade",
        "Backtest": True,
        "Live": True,
        "Research": True,
        "TradFi": True,
        "CeFi": True,
        "DeFi": False,
        "Prediction Market": False,
    },
    "finrobot": {
        "github": "ai4finance-foundation/finrobot",
        "Status": ProjectStatus.ACTIVE,
        "Approach": None,
        "Backtest": False,
        "Live": False,
        "Research": True,
        "TradFi": None,
        "CeFi": None,
        "DeFi": None,
        "Prediction Market": None,
    },
    "trading_agents": {
        "github": "tauricresearch/tradingagents",
        "Status": ProjectStatus.ACTIVE,
        "Approach": None,
        "Backtest": False,
        "Live": False,
        "Research": True,
        "TradFi": None,
        "CeFi": None,
        "DeFi": None,
        "Prediction Market": None,
    },
    "ai-hedge-fund": {
        "github": "virattt/ai-hedge-fund",
        "Status": ProjectStatus.ACTIVE,
        "Approach": None,
        "Backtest": False,
        "Live": False,
        "Research": True,
        "TradFi": None,
        "CeFi": None,
        "DeFi": None,
        "Prediction Market": None,
    },
    "dexter": {
        "github": "virattt/dexter",
        "Status": ProjectStatus.ACTIVE,
        "Approach": None,
        "Backtest": False,
        "Live": False,
        "Research": True,
        "TradFi": None,
        "CeFi": None,
        "DeFi": None,
        "Prediction Market": None,
    },
}
repos_table(REPOS)

“✔️” = SUPPORTED
“✖️” = NOT SUPPORTED
“➖” = NOT APPLICABLE

💡 Concepts Explained

Code-to-Trade Approach¶

This means asking an LLM to write Python code, either using a chosen algo-trading framework or writing ad-hoc code, so that a strategy can be backtested, traded deterministically, with reproducible results. A good visual example is QuantConnect’s agents (video).

Prompt-to-Trade Approach¶

This means giving a prompt to an LLM and letting it make trading decisions non-deterministically, meaning the same input may produce different results across runs. In other words, “Vibe-Trading”. Think of it as OpenClaw with trading ability. A good visual example is Alpha Arena, where agents trade based on prompts.

Running LLMs for free

In order to run the AI apps, you will need access to LLMs.

LLM Providers (with generous free-tier limits):

Running Local LLMs:

Remarks

The creator of AI-Hedge-Fund and Dexter is also the founder of Financial Datasets, the financial data provider recommended in the projects’ READMEs, but this affiliation is not clearly disclosed.
AI-Trader is another project from the same organization behind Vibe-Trading, serving as a prompt-to-trade AI trading platform for AI agents such as OpenClaw, nanobot.

Agent Frameworks¶

REPOS = {
    "pydnatic-ai": {
        "github": "pydantic/pydantic-ai",
        "Status": ProjectStatus.ACTIVE
    },
    "smolagents": {
        "github": "huggingface/smolagents",
        "Status": ProjectStatus.ACTIVE
    },
    "openai-agents": {
        "github": "openai/openai-agents-python",
        "Status": ProjectStatus.ACTIVE
    },
    "agno": {
        "github": "agno-agi/agno",
        "Status": ProjectStatus.ACTIVE
    },
    "crewai": {
        "github": "crewaiinc/crewai",
        "Status": ProjectStatus.ACTIVE
    },
    "deepagents": {
        "github": "langchain-ai/deepagents",
        "Status": ProjectStatus.ACTIVE
    },
    "ag2": {
        "github": "ag2ai/ag2",
        "Status": ProjectStatus.ACTIVE
    },
    "autogen": {
        "github": "microsoft/autogen",
        "Status": ProjectStatus.INACTIVE
    },
    
}
repos_table(REPOS)

LLM Frameworks¶

REPOS = {
    "haystack": {
        "github": "deepset-ai/haystack",
        "Status": ProjectStatus.ACTIVE,
    },
    "langchain": {
        "github": "langchain-ai/langchain",
        "Status": ProjectStatus.ACTIVE,
    },
    "llamaindex": {
        "github": "run-llama/llama_index",
        "Status": ProjectStatus.ACTIVE,
    },
}
repos_table(REPOS)

Agent vs. LLM frameworks — a fuzzy line

These two groups overlap, so don’t read the split as rigid. Many tools do both — langchain, for instance, can create agents but is centered on LLM applications. The grouping reflects each framework’s primary focus, not its full capability:

LLM frameworks focus on connecting models to your data — retrieval (RAG), and ingesting/parsing/indexing documents into something an LLM can use (e.g. llamaindex).
Agent frameworks focus on agentic workflows — giving the model tools, memory, and multi-step autonomy to act.