If you've been looking for a financial agent that genuinely does the dirty work of research for you — rather than yet another chat wrapper — DojoAgents is worth ten minutes of your time. It's the personal-investing agent framework inside "Autonomous Finance Intelligence AlphaDojo," open-sourced by Shenchong Intelligence (Alpha-Dojo); it collected more than 1,000 GitHub stars in just over a week.

It breaks market scanning, news retrieval, company analysis, industry-chain mapping, portfolio construction, and risk analysis into a set of tools the agent can call autonomously, then chains them into a complete decision-support workflow via an Agent Loop, covering A-shares, US stocks, and Hong Kong stocks. This tutorial walks you through running it locally.

⚠️ Important prerequisite: the framework itself ships with no licensed market-data or news sources; you must configure authorized data providers yourself. The project carries a strong regulatory disclaimer — for information organization and research only, and not investment advice of any kind.

What You Need Before Starting

• Python 3.11+ (in practice, 3.11 or higher)
• uv (package manager; less hassle than pip)
• A working LLM key (pick one of OpenAI / Anthropic / Gemini / DeepSeek / GLM; local models are also supported)
• Estimated time: 10-15 minutes

Project repository: https://github.com/Alpha-Dojo/DojoAgents (PyPI package name: dojoagents)

Step 1: Create a Virtual Environment and Install

Open a Terminal and run the following in your working directory:

uv venv && source .venv/bin/activate
uv pip install dojoagents

If you don't have uv yet, run brew install uv first (macOS) or see the official uv installation docs.

Step 2: Start the Dashboard

Once installation finishes, one command brings up the local console:

dojoagents dashboard --host 127.0.0.1 --port 8765

Then open http://127.0.0.1:8765/ in your browser.

Step 3: Configure Models in Settings

On the Settings page you can configure the models you want the agent to use. Two types of integration are supported:

• API models: OpenAI, Anthropic, Gemini, DeepSeek, GLM
• Local models: Ollama, vLLM, llama.cpp

If you want to save money or run offline, take the local-model route; if you want reasoning quality, take closed-source APIs like Anthropic / DeepSeek / GLM.

After the Four Typical Scenarios Are Running

Once the dashboard is up, the four scenarios below are where the project demos show the Agent Loop at its best — try them in order.

Scenario 1: Market Theme Scan

Ask the agent directly: "What sectors are moving and which names are strong today across A-shares, US stocks, and Hong Kong stocks?"

The agent calls its global-market, industry, and individual-stock tools, pulls movers, price changes, volume shifts, and sector distribution, then groups them into themes like semiconductors, AI compute, storage, and optical modules. You can follow up with "turn this into a research watchlist," and it will output a structured list of names.

Scenario 2: From News to Opportunity Mining

Try a question with a causal chain in it: "Meta is selling compute — which US and A-share industries and companies does that affect?"

The agent's processing path is worth watching:

1. First pull Meta's price and related news
2. Analyze the event's impact path (AI compute → cloud computing → servers → GPU → optical modules → semiconductor equipment)
3. Map it onto specific US and A-share companies
4. Optionally, build a simulated watch portfolio based on prices as of the news date, with simulated entry prices and a simulated return curve displayed in the dashboard

💡 Tip: the simulated portfolio is for research purposes, not a backtested promise. Historical price analysis only covers the period for which the agent can obtain data.

Scenario 3: Anonymous Diagnosis from a Portfolio Screenshot

This is the scenario that best shows off its multimodal side. Upload a screenshot of your holdings with your name and account masked (the demo uses 40+ names across A-shares, US stocks, and ETFs). DojoAgents will:

• Re-classify holdings by market, sector, and industry-chain position
• Generate a portfolio risk diagnosis
• Flag issues like over-concentration in one theme, heavy internal overlap between holdings, and insufficient defensive assets

Always redact before uploading — the agent doesn't need to know who you are.

Scenario 4: Market Discovery Aggregation

It automatically aggregates multiple news sources, institutional views, and market movers, distills the day's dominant market themes, and then analyzes the specific industries and sectors affected. Good for a morning-briefing run before each market open.

Why It's Worth Your Time

Autonomous finance intelligence is still early. Many teams keep rebuilding the same underlying stack: model adapters, data interfaces, tool calling, task orchestration, frontend interaction. What AlphaDojo opens up is that base framework: Agent Loop, task orchestration, financial tool interfaces, dashboard, multi-model adaptation, local deployment, and portfolio research capabilities.

If you're a financial developer or quant researcher, you can swap in your own models, extend the tools, and plug in your own data and knowledge base — saving the time of building a framework from scratch.

Common Issues

• Startup error "module not found": make sure the virtual environment is activated (source .venv/bin/activate) and Python is ≥ 3.11.
• Market data comes back empty: the framework ships without data sources; you need to configure authorized ones yourself. The open-source repo's docs list the supported adapters.
• Local Ollama model returns nothing: check that the Ollama service is reachable at http://localhost:11434 and that the model has actually been pulled with ollama pull.
• A-share data is inaccurate: A-share data quality depends entirely on the data source you connect — never treat the agent's output directly as a trading signal.

Disclaimer: DojoAgents is intended mainly for organizing financial information, research analysis, and technical demonstration, and does not constitute investment advice of any kind; portfolio construction and historical price analysis are simulated research scenarios.