Dexter

Dexter is an autonomous agent purpose-built for financial research. It uses an agentic loop that plans, executes, validates, and iterates on analysis tasks. When given a complex financial question, Dexter automatically decomposes it into structured research steps. It selects and executes the right tools to gather financial data including income statements, balance sheets, and cash flow statements from the Financial Datasets API. The agent checks its own work and iterates until tasks are complete, using self-reflection to improve accuracy. Built with TypeScript and Bun runtime, Dexter supports multiple LLM providers including OpenAI, Anthropic, Google, xAI, OpenRouter, and local models via Ollama. This flexibility means you can use the cheapest or most capable model for each specific task. It includes built-in safety features like loop detection and step limits to prevent runaway execution. Dexter has 22,700+ GitHub stars and 2,700+ forks, making it one of the most popular open-source AI agent projects. The project includes an evaluation suite with LangSmith integration for tracking research accuracy. Each query creates a scratchpad log file for full traceability. Key capabilities include intelligent task planning that decomposes complex questions, autonomous execution with the right financial tools, self-validation through iterative refinement, real-time access to corporate financial data, and comprehensive safety features with explicit step limits and loop detection. Users can run Dexter interactively in the terminal or via WhatsApp for on-the-go financial research. The agent supports web search through Exa or Tavily APIs for supplementary data gathering, and financial data through the Financial Datasets API for institutional-grade market data. Installation requires Bun runtime, an OpenAI API key, and a Financial Datasets API key. The project is open source with a thriving community on Discord. Dexter represents a practical application of autonomous agent architectures for real-world financial analysis tasks.