“"Think Claude Code, but built specifically for financial research." — Dexter README”
You know that feeling when a finance question sends you through filings, price data, web search, and a pile of notes before you can write one clear answer. Generic chat tools can help, but they do not start with finance tools, and they do not keep a clean trace of each step by default. Dexter targets that mess by giving you one CLI flow that plans the job, runs the tools, and keeps a scratchpad on disk. You spend less time stitching sources together by hand.
Think of Dexter like a junior analyst with a checklist and a notebook. You start it in your terminal and ask a research question, then it breaks the job into steps and calls finance tools, web search, browser actions, and skills such as DCF work. As it works, it writes tool results into a JSONL scratchpad and streams events through the CLI, so you can inspect the path instead of trusting a black box. When it has enough evidence, it makes a separate final-answer call and writes the response.
This fits you if you build AI tools, work on fintech products, or spend time turning finance questions into repeatable research workflows. It also fits if you want to study how a terminal agent mixes domain tools, browser actions, and a scratchpad into one loop. It is not a strong fit yet if you need published benchmark scores, a fully self-contained local install, or an offline setup with no API keys.
Yes, if you care about finance-specific agent design and you want real repo activity instead of a static demo. The release cadence, 20 contributors, and hosted paid product push it past the toy stage, but the missing public benchmark data and the multi-key setup keep it in beta territory for me. Treat it as a serious base to study or extend, not as a drop-in answer engine you trust blindly.
Deep-dive insight, Easy and Pro modes, plus action playbooks — the full breakdown is one tap away.