Dexter is an autonomous financial research agent built in TypeScript that performs deep analysis of complex financial questions through task planning, self-reflection, and real-time market data integration. The agent decomposes intricate financial queries into structured research steps, executes those tasks using live financial data sources, validates its own work iteratively, and refines results until reaching confident, data-backed answers. The project explicitly positions itself as educational and informational only, with clear disclaimers that it is not intended for real trading or investment decisions.
The core capabilities of Dexter center on intelligent task planning that automatically breaks down complex queries into actionable research steps, autonomous execution that selects and runs appropriate tools to gather financial data, and self-validation mechanisms that check work and iterate until tasks complete successfully. The agent has access to real-time financial data including income statements, balance sheets, and cash flow statements. Safety features are built in, including loop detection and step limits to prevent runaway execution. The system is designed to function similarly to Claude Code but specifically tailored for financial research workflows.
The technical foundation uses Bun as the runtime environment and requires API keys from OpenAI, Financial Datasets, and optionally Exa for web search capabilities. Installation involves cloning the repository and installing dependencies through Bun, with environment variables configured for API access. Users can run Dexter in interactive mode or with watch mode for development purposes.
The repository includes a comprehensive evaluation suite that tests the agent against a dataset of financial questions using LangSmith for tracking and an LLM-as-judge approach for scoring correctness. Developers can run evaluations across all questions or on random samples, with a real-time UI displaying progress, current questions, and running accuracy statistics. All tool calls are logged to a scratchpad file system located in the .dexter/scratchpad/ directory, with each query creating a new JSONL file containing newline-delimited JSON entries that track the original query, tool results with arguments and summaries, and agent reasoning steps.
The project supports WhatsApp integration through a gateway system, allowing users to chat with Dexter directly through WhatsApp by messaging themselves. Responses are processed by the agent and sent back to the same chat conversation.
According to GitGenius activity tracking, the repository shows a median issue and pull request response latency of 2.7 hours across 76 tracked items, with a mean latency of 88.1 hours. The most active issue labels are help wanted with 10 occurrences, enhancement with 3, and good first issue with 1. Primary contributor activity is dominated by virattt with 127 tracked events, followed by bittoby and claytonlin1110 each with 10 events. The repository shares overlapping contributors with crewaiinc/crewai, microsoft/autogen, and openclaw/openclaw, indicating connections to broader autonomous agent and AI research communities. GitGenius classifies the repository across multiple domains including static sites, markdown, HTML generation, templating, website builders, content publishing, Python, web content, generators, and CLI utilities, reflecting its diverse technical scope and tooling ecosystem.
