- Category
- AI Agents
- Rank
- No. 188Tools index
- Type
- APP
- Builder
- virattt
- GitHub
- 27.6k stars
- Latest release
- v1.0.5
- Date
About
Dexter is an autonomous financial-research agent that thinks, plans, and learns as it works, using task planning, self-reflection, and real-time market data — think Claude Code, but built specifically for financial research.
What it does
You ask a financial question in a terminal or through a linked WhatsApp self-chat. Dexter builds a research sequence, chooses data or search tools, gathers company statements and market information, checks progress, and revises its response. Each query also produces a structured trace of reasoning, tool inputs, raw results, and summaries for later inspection.
Why it's ranked here
Dexter combines a usable terminal interface with unusually visible agent operations. It supports several hosted model providers and local Ollama, records detailed query traces, and includes loop detection, step limits, and a finance-specific evaluation runner. That makes it credible for experimentation and debugging. Its own disclaimer, external data dependency, and model-judged evaluation keep the verdict measured.
What's good
The provider registry centralizes model routing, credentials, fast-model choices, and context limits. The interface exposes thinking, tool progress, approvals, interruptions, token usage, and completion statistics. Retry handling distinguishes errors that should not be retried. Evaluation supports timeouts, concurrent runs, seeded samples, rubric criteria, contradiction detection, and separate agent and judge latency tracking.
Tradeoffs
Setup requires Bun plus credentials for a model provider and the financial-data service. Web research needs another provider, though it is optional. Installation also downloads Chromium through Playwright. Outputs may be wrong, incomplete, or stale, and the project explicitly excludes real trading, investment advice, tax advice, and legal advice. Evaluation correctness depends on an LLM judge.
How to use it well
Use Dexter for exploratory company research when you want a conversational terminal workflow and an inspectable record of every data-gathering step. It best suits developers or financially literate researchers who can verify claims against primary material. Run its evaluation suite when changing models or agent behavior. Do not treat it as an execution system, portfolio manager, or substitute for professional advice.
Technical notes+
package.json defines an ESM TypeScript package named dexter-ts, exposes the dexter-ts binary, runs through Bun, and installs Chromium in postinstall. src/index.tsx loads environment variables and calls runCli() from src/cli.ts, where a persistent TUI component tree renders queries, tool state, approvals, questions, interruptions, answers, and performance data. src/providers.ts is the canonical provider registry. src/model/llm.ts routes models by prefix, configures hosted and Ollama backends, binds tools, extracts usage, applies retries, and adds Anthropic prompt caching. src/evals/run.ts runs finance questions with timeouts and concurrency, applies rubric judging, and records experiments through LangSmith.
Observed
- License
- MIT License
- Primary language
- TypeScript
- Packaging and install
- ESM package installed from a cloned repository with Bun; installation downloads Playwright Chromium.
- Interfaces
- Interactive terminal CLI and a WhatsApp gateway for self-chat messages.
- Model support
- Registry includes OpenAI, Anthropic, Google, xAI, Moonshot, DeepSeek, OpenRouter, Ollama, and Ollama Cloud.
- Data and search dependencies
- Financial Datasets supplies market data; Exa is preferred for web search with Tavily as fallback.
- Evaluation surface
- Finance-question runner supports sampling, timeouts, concurrency, rubric judging, and LangSmith tracking.
- Documented platform setup
- Bun installation instructions cover macOS, Linux, and Windows.
Read from README.md, package.json, src/cli.ts, src/index.tsx, src/theme.ts, src/types.ts, src/providers.ts, src/evals/run.ts, src/model/llm.ts.
What it can do
Decompose complex financial questions into structured research plans
Complex financial query or question → Step-by-step research plan with specific tasks
Retrieve real-time financial statements and market data
Company ticker or financial data request → Income statements, balance sheets, and cash flow statements
Execute autonomous financial analysis tasks
Research plan and data requirements → Completed financial analysis with gathered data
Validate and refine research results through self-reflection
Initial analysis results and research objectives → Refined and validated financial research findings
Generate data-backed financial research reports
Financial questions and market data → Comprehensive research report with confident conclusions
Detect and prevent infinite analysis loops
Analysis execution state and step count → Safe termination or continuation decision
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.