Automate Financial Research with the OpenClaw of Stocks

In the demo, we used Hostinger VPS hosting and picked a basic VPS plan. Once the server is ready, Hostinger gives you an SSH command that looks like this:

ssh root@[YOUR_SERVER_IP]

> Pro tip: You can run Dexter locally, but a VPS is cleaner for this kind of agent. It keeps the browser, terminal, and long-running tasks off your main laptop. If setting up a VPS is new for you, start with our Launch Your OpenClaw From Scratch course. It walks through the same basic server setup pattern you will use here.

Open a new Codex or Claude Code project, open the terminal, and connect to the server with that SSH command. The nice part is that your coding agent can watch the terminal with you and help debug installs, missing packages, or environment variables.

In the demo, AlmaLinux worked, but Debian would be the safer pick next time.

Open the Dexter GitHub repo and follow the install instructions there. The setup is mostly a few terminal commands: install Bun, clone the repo, install dependencies, add environment variables, and start Dexter.

The general flow looks like this:

# install Bun if you do not already have it
curl -fsSL https://bun.sh/install | bash

# clone Dexter and install dependencies
git clone https://github.com/virattt/dexter.git
cd dexter
bun install

# start Dexter
bun start

When Dexter starts for the first time, it should walk you through the variables it needs. Treat those variables like passwords. Do not paste them into a public recording, shared doc, or screenshot.

After setup, the command you will use most often is:

cd dexter
bun start

That reopens the Dexter interface from the folder where it lives on your VPS.

Now give Dexter the two things that make it useful: a model and stock-market data.

For the model, use whichever API key you already prefer: OpenRouter, OpenAI, or Anthropic. In the demo, OpenRouter was useful because it can route to OpenAI, Anthropic, and other models from one account.

Inside Dexter, open the model picker with:

/models

Choose your provider, then paste or type the model name carefully. In the demo, straight pasting a model name caused issues, so typing it out was the safer move.

Next, add a Financial Datasets API key. This is the key that lets Dexter pull company data, earnings reports, SEC filings, stock prices, and other financial context instead of relying only on web search.

Financial Datasets charges per use, and the demo notes that the minimum credit load is about $20 to get started.

If you want web search, add a Perplexity API key and configure search inside Dexter with:

/search

If you want Twitter/X sentiment later, add an xAI API key too.

Start with a narrow research question. Dexter works better when you give it a concrete company and decision context instead of asking it to "tell me about this stock."

Use this prompt:

Create a source-backed research brief on GOOGL.

Focus on:
- what the company does
- recent catalysts
- public financial or operating signals
- key risks
- competitive context
- what I should investigate next

Use an iterative research process.
Validate important claims before finalizing the brief.
Save intermediate research so the workflow can continue without losing context.

You can also ask faster questions while testing the setup:

What's the stock price of GOOGL?

or:

How did GOOGL do on its last earnings? How did that compare to META's last earnings?

In the demo, Dexter pulled market data, read filing and earnings fields, then returned a quick comparison of revenue, EPS, and growth. That is the workflow you want to verify before trusting bigger research prompts: ask a concrete question, watch which data sources Dexter calls, and make sure the answer cites the right kind of evidence.

Once a manual run works, use Dexter's heartbeat feature for recurring checks.

Use a prompt like this:

Set up a heartbeat to check [TICKER] every [TIME_WINDOW].

When you check it, report:
- current price movement
- any major earnings or filing updates
- relevant market news
- whether anything needs deeper research

In the demo, the heartbeat request was a simple six-hour stock price check. Dexter added it to its heartbeat checklist so the agent could keep checking that stock on a schedule.

This is where Dexter starts to feel different from a normal chatbot. You are not opening a new thread every time you want a quick check. You are giving a single-purpose finance agent one recurring job.

> Pro tip: Prove the manual workflow before you automate it. If the first GOOGL answer is noisy, fix the model, data, or prompt before you ask Dexter to monitor anything.

Dexter also has gateway instructions for WhatsApp-style access, but that part did not work cleanly in the demo. The setup required exiting Dexter with Ctrl+C, returning to the normal terminal, and trying:

bun gateway login

The gateway setup had more friction than the core stock research workflow. Treat that as optional for now.

The stronger first use is simple:

1. Run Dexter on a VPS. 2. Add the right model and market data keys. 3. Ask source-backed questions. 4. Check the data sources it used. 5. Add /heartbeat only after the answers are useful.

That keeps Dexter focused on what it is already good at: stock research, market data, filings, earnings context, and recurring watchlist checks.