Agents Demo

A small, opinionated template for building an “agentic” crypto research + trading assistant:

• Tavily: web search tool for news/market context
• Alpaca: paper trading (account, assets, positions, orders)
• PydanticAI: enforces structured outputs (e.g., buy/sell/hold + confidence) and tool calling
• State file: minimal persistence so the “runner” can reconcile what happened after placing an order

This is intentionally minimal - feel free to fork or suggest PRs to add to existing codebase. This acts more as an experiment to let the model 'think' for itself and reason what to do next whilst providing the necessary tools to act on it.

NOTE: To change the model use, navigate to main.py - most models are supported from Ollama to OpenAI. If it is supported by PydanticAI, it can be used.

NOTE: To change level of logging, simply navigate to utils/logging.py and change to logging.INFO for all logs.

---

Setup

• Python 3.11+

• uv (recommended) or pip

• make sure to run uv sync then run the main.py via the command uv run python main.py

• Or can just use Docker setup too to run via the command docker compose up --build -d

• To see docker logs in terminal (not just Docker dashboard) run docker compose logs -f agent

• NOTE: Need to create an env file that imitates the following pattern, be sure to get the corresponding keys and secrets from corresponding sources:

  TAVILY_API_KEY=your_tavily_key_here
  ALPACA_KEY=your_alpaca_key_id_here
  ALPACA_SECRET=your_alpaca_secret_key_here
  ALPACA_URL=https://paper-api.alpaca.markets/v2
  ANTHROPIC_API_KEY= your_anthropic_key_here