Dexter Review in 2026: The Open Source Financial Research Agent
8 min read

Dexter (virattt/dexter) is a free, open source TypeScript agent that plans financial research, pulls public company data through the Financial Datasets API, and checks its own work. It is a strong learning and prototyping tool for listed US equities, but its README says it is not for real trading, the repo has no LICENSE file despite claiming MIT, and any fund use needs Financial Datasets' $1,000 per month commercial license.
What this covers
- Dexter is an autonomous financial research agent in TypeScript on Bun, with 27,842 GitHub stars as of 16 November 2026.
- Its finance tools cover statements, prices, news, insider trades, SEC filings and a stock screener, all through the Financial Datasets API.
- The README claims MIT, but the repo has no LICENSE file and GitHub reports the license as null.
- Financial Datasets requires its Professional plan, $1,000 per month, for any company or fund use.
- The README says Dexter is for educational and informational use and not for real trading or investment.
- Private deal work runs on data rooms and management accounts, which Dexter's data tools do not reach.
Dexter is an open source, autonomous agent for financial research, built by Virat Singh (GitHub handle virattt), that breaks a finance question into research steps, pulls company data through tools, checks its own work, and returns an answer in your terminal. It is a strong research and prototyping tool for public US equities, but it is not a production system for a fund: its own README says it is not intended for real trading or investment, and its data source and license both need checking before any commercial use.
This review covers what Dexter is, what it does in practice, the license question, what it costs to run, and where it fits for a fund, family office or deal team. Every product fact below comes from the Dexter GitHub repository, the Financial Datasets pricing page, or an independent write-up, each fetched on 2026-11-16.
What Dexter is
The project describes itself as "an autonomous financial research agent that thinks, plans, and learns as it works," and pitches itself as "Think Claude Code, but built specifically for financial research." It is written in TypeScript and runs on the Bun runtime. You clone the repo, add API keys to a .env file, and run it as an interactive command line app.
As of 2026-11-16, the repository shows 27,842 stars, 3,435 forks and 159 watchers on GitHub. It was created on 14 October 2025 and the most recent push was on 23 September 2026. The package version is 1.0.6. The contributor list is concentrated: the creator accounts for 445 commits, and the next two contributors have 10 and 6.
Star counts measure attention more than use. An independent review on Stackademic in February 2026 made the same point, noting a very high ratio of stars to watchers at the time. Treat the star count as evidence that people are curious about agentic finance research, not as evidence of production adoption.

AI for finance teams
Screen more deals without adding analysts
First system live in about six weeks. You own the code.
Book a discovery callWhat it actually does
The README lists five headline capabilities: task planning that decomposes a question into research steps, autonomous tool selection, self-validation that iterates until a task is complete, real-time financial data covering income statements, balance sheets and cash flow statements, and safety features such as loop detection and step limits. In the source, the default cap is 10 iterations per query, after which the agent stops and says it could not finish.
The tool registry in the code shows what the agent can reach. The finance tools are get_financials (statements and metrics), get_market_data (stock and crypto prices, company news, insider trades and ownership), read_filings (extracts and summarizes sections of 10-K, 10-Q and 8-K filings) and stock_screener (screens by P/E, growth, margins and similar). Around those sit general agent tools: web search, web fetch, a headless browser, file read and write, a bash tool, memory, scheduled jobs, and the ability to spawn subagents that run sub-tasks in parallel.
Dexter also ships "skills," which are scripted workflows the agent follows. The repo includes a DCF valuation skill, an investment memo skill that produces a one-page HTML memo "structured the way a hedge fund analyst would present a thesis to a PM," and an X research skill. The Stackademic reviewer singled out the DCF skill as the strongest part of the project, describing it as a multi-step workflow with revenue projection, margin analysis, discount rate, terminal value and a 3x3 sensitivity table.
Two engineering details matter to anyone evaluating it seriously. First, every query writes a JSONL scratchpad file under .dexter/scratchpad/ recording the original question, each tool call with its arguments and raw result, and the agent's reasoning steps. That is a useful audit trail. Second, the repo includes an evaluation suite that runs the agent against a dataset of financial questions, tracks runs in LangSmith, and scores answers with an LLM-as-judge approach. Both are signs of a project built by someone who has thought about trust in agent output.
There is also an optional WhatsApp gateway, so you can message the agent from your phone. It is convenient for a personal setup and a governance question for a firm, since research queries would leave through a consumer messaging channel.
The license, and what it allows commercially
The README states: "This project is licensed under the MIT License." MIT is a permissive license that normally allows commercial use, modification and private deployment, provided the copyright notice is kept.
The complication is that the repository root, as fetched on 2026-11-16, contains no LICENSE file, and GitHub's API reports the license field as null. The Stackademic review flagged the same gap in February 2026 and argued that, without a license file, the legal default is all rights reserved. It is most likely an oversight. For a hobbyist it changes nothing. For a fund that wants to fork the code and run it inside its own environment, it is a question for counsel, and the cheapest fix is asking the maintainer to add the file or confirm the grant in writing.
The code license is also only half of the commercial picture. The data license is the other half, and it is the one more likely to bite.
What it costs to run
Dexter itself is free to download. Running it means paying for three things: a language model, financial data, and optionally web search. The README lists an OpenAI API key and a Financial Datasets API key as prerequisites, with an Exa key optional for web search. The example environment file also supports Anthropic, Google, xAI, OpenRouter, Moonshot and DeepSeek models, plus Ollama for local models, and Perplexity, Tavily and LangSearch as fallback search providers.
Infrastructure is minimal. It runs on a laptop. Note that the install step runs playwright install chromium, so the first install downloads a browser for the agent's browsing tool.
The financial data comes from Financial Datasets, and every finance tool in the repo calls that API. According to the Stackademic review, Financial Datasets is the creator's own company, which the reviewer framed as an open source project that doubles as distribution for a paid API. That is a common and legitimate model, but it explains the dependency. The live pricing page, fetched on 2026-11-16, shows:
| Financial Datasets plan | Price | Included requests | License |
|---|---|---|---|
| Credits | $20 one-time | 1,000 | Personal |
| Personal | $200 per month | 100,000 | Personal |
| Professional | $1,000 per month | 1,000,000 | Commercial, with redistribution rights, webhooks and uptime SLAs |
| Enterprise | Custom | Custom | Custom contracts, zero data retention, bulk delivery |
The licensing line is the one a firm must read. The pricing page says the Personal license is for individuals using the data for "personal, non-business purposes" and states: "If you use the API for a company, a fund, or any work that benefits others, you need Professional." So for any fund, family office or deal team, the realistic floor for the data alone is $1,000 per month. Overage on Professional is $5 per 1,000 requests, and some premium datasets, such as forward guidance and operating KPIs, count as multiple requests per call.
Model cost depends on the model you pick and how many tool calls each question triggers. Financial Datasets reports coverage of 27,530 tickers, with annual fundamentals for UK and French companies as the international extension, so the data is overwhelmingly a public equities set.

Build library
See what we have already shipped
Working systems, the stack behind them, and what they cost to run.
Open the libraryWho should use it, and who should not
Dexter is a good fit for an analyst or engineer who wants to learn how a finance research agent is built, a small team prototyping an internal research assistant, or an individual investor doing fundamental work on listed US companies. The scratchpad, the eval harness and the DCF and memo skills are worth reading even if you never run it. The codebase is a clear reference implementation of a planning agent with tool routing.
It is a weaker fit where the work is private. Dexter's data tools cover listed companies, filings, prices and ownership. A private equity diligence, a family office reviewing a direct deal, or a credit team underwriting a private borrower works from a data room, management accounts and a quality of earnings report. None of that sits behind the Financial Datasets API. You would be rebuilding the tool layer, not configuring it.
It is also not a trading or backtesting system. The README is explicit that it is for "educational, entertainment, and informational purposes only" and "not intended for real trading or investment." The independent review adds that it cannot trade and does not backtest. If you need a strategy tested against history before money moves, Dexter is not that tool.
Finally, the bus factor is real. One person wrote the large majority of the commits. That is fine for a reference project and a risk for a dependency in a regulated workflow.
Where it fits for a fund, family office or deal team
The honest framing is research tool versus production system. Dexter shows that a planning agent can pull statements, read a 10-K, run a structured DCF and draft a memo with a traceable log. A production system for a firm has to do more: run inside the firm's own environment, read the firm's own documents, respect data licenses and retention rules, keep an audit trail a compliance officer will accept, and be maintained by someone accountable to the firm.
A sensible path for a team curious about Dexter looks like this:
- Run it on a laptop with a test budget to see how it reasons on companies you know well.
- Read the scratchpad files and check every number against the source filing. That tells you more than any demo.
- Before any firm use, settle the code license with the maintainer and move to a commercial data license.
- Decide whether your real work is public equities research or private deal work. The answer decides whether Dexter is a starting point or only an inspiration.
For private markets work specifically, see how we approach AI for financial due diligence and AI for private equity firms, where the inputs are data rooms and management packs rather than a market data API.
When an in-environment build is the better answer
If your workflow runs on private documents, needs to stay inside your own cloud, and has to survive a compliance review, an open source agent wired to a third party data API is usually the wrong base. The agent pattern Dexter demonstrates is sound. The tools, data access, controls and ownership are what need to be built for your firm.
That is what consultance.ai does. We build custom finance AI systems inside the client's environment, with the first system typically live in about six weeks, and the client owns the code. If you want to see what that looks like for a deal desk, start with our AI deal desk systems, browse the free prompt packs in the library, or book a call to talk through your workflow.

Implementation services
Put AI into the workflow you actually run
Diligence, close, reconciliation, reporting — in your environment.
See how it worksSources
- virattt/dexter on GitHub: README, env.example, package.json, src/tools/registry.ts, src/agent/agent.ts, src/skills (fetched 2026-11-16)
- GitHub API repository metadata for virattt/dexter: stars, forks, watchers, license field, dates, contributors (fetched 2026-11-16)
- Financial Datasets pricing and license terms (fetched 2026-11-16)
- JP Caparas, "Is Dexter the 'Claude Code for finance' it claims to be?", Stackademic, 8 February 2026 (fetched 2026-11-16)
