easy_investment_Agent_crewai: A-Share Analysis with Four CrewAI Agents and AKShare

What easy_investment_Agent_crewai Actually Solves

Most retail A-share research ends up as four browser tabs: a quote page, a financial statement page, a capital-flow page, and a news feed. The project's premise is that this assembly work is the boring part, and that a language model with tool access can do it in one pass. It targets people who already know what a 资金流向 table looks like and want the gathering automated, not people who want a signal handed to them.

The README is explicit about the boundary: the system is for study and research and does not constitute investment advice. That sentence is not boilerplate here, because nothing in the repository places orders, tracks positions, or backtests a strategy. The output is prose analysis from four agents plus whatever the configured model writes into the final task. If your actual need is execution or portfolio accounting, this is the wrong tool and no amount of configuration will change that.

The second audience is CrewAI users looking for a non-toy example. Four roles with distinct tools and a shared data source is a more realistic shape than the single-agent demos in most tutorials, and the A-share domain adds constraints (price limits, policy sensitivity, retail sentiment) that a generic equity agent ignores.

Four Agents, Four Tool Groups, One Shared Data Layer

The architecture is a standard CrewAI crew with a domain-specific split. The README names four roles: an A-share market analyst covering technical, policy and capital-flow angles; a financial statement specialist doing ratios, trend analysis and peer comparison; a market sentiment researcher on capital flows, news tone and policy impact; and an investment advisor who synthesizes and writes strategy and risk control.

Each role draws on a matching tool group. The data tools wrap AKShare for real-time quotes, historical K-lines, financial indicators, capital flow, sector data and sentiment indicators. A separate financial analysis tool set handles deeper ratio work and peer comparison, a sentiment tool set covers capital flow and news tone, and a calculator tool exists so the model does not do arithmetic in its head. That last one is the detail that tells you the author has watched an LLM get a percentage wrong.

The data flow is linear and worth understanding before you customize anything. Tasks are declared in config/tasks.yaml, agents in config/agents.yaml, and the crew in src/a_stock_analysis/crew.py binds them to a model. Inputs are a company name, a stock code and a market, passed as a dictionary. Everything downstream is the model reading tool output and writing prose. There is no caching layer, no database, and no persistence between runs mentioned anywhere in the README, so each run re-fetches from AKShare.

Installing It and Running Your First A-Share Analysis

The README requires Python 3.12 or newer and recommends Poetry. Clone the repository, then install dependencies. The --no-root flag is in the README and matters: without it Poetry tries to install the project itself as a package.

poetry install --no-root

The README also gives a UV path, which is the shorter route if you already use it. uv sync creates the virtual environment and installs everything in one step.

uv sync

Next, copy the environment template and fill it in. The README does not enumerate which variables are required, so read .env.example before assuming an API key is enough.

cp .env.example .env

Now run the default analysis. The README states the system defaults to Tencent Holdings (00700.HK).

poetry run python src/a_stock_analysis/main.py

There is also an installed entry point, which is the cleaner way to invoke it repeatedly.

poetry run a_stock_analysis

To analyze a mainland stock instead, edit the inputs dictionary in main.py. The README gives this exact example for Kweichow Moutai.

inputs = {
    'company_name': '贵州茅台',
    'stock_code': '600519.SH',
    'market': 'SH'
}

Stock code suffixes follow the exchange: 600519.SH for Shanghai, 000001.SZ for Shenzhen, 00700.HK for Hong Kong. Expect a long wait on the first run, since four agents each making AKShare calls is not fast, and expect the quality of the output to track whichever model you configured rather than the framework.

Where the Scaffold Breaks Down

The default model is Ollama running llama3.1 locally, per the crew.py example in the README. A local 8B-class model writing four-stage financial analysis is the weakest link in the chain, and the README does not discuss how output quality changes when you switch to a hosted model. It just shows the ChatOpenAI swap. Anyone evaluating this should treat the model choice as the primary quality variable, not the agent design.

AKShare is the second failure surface. It scrapes public Chinese financial sources, and those endpoints change. The README lists the data categories the tools cover but says nothing about error handling when a source returns empty or restructured data. A run that silently produces analysis from a failed fetch is worse than a run that crashes, and nothing in the README suggests the former is guarded against.

The repository layout itself is a signal. The top level contains README.md, stock_analysis/ and stock_analysis_a_stock/, while the README's install steps cd into stock_analysis_a_stock. Two similarly named directories with no explanation in the README means you should confirm which one the entry point actually imports before you start editing files. That is a five-minute check that saves an hour.

Finally, the licence. The README says MIT. There is no licence identifier in the repository metadata, and no LICENSE file is listed in the top-level entries. MIT in a README is a statement of intent, not the same as a licence file that a legal reviewer can point at.

How It Differs from Hand-Wiring CrewAI Yourself

The honest alternative is not another A-share product. It is building the same crew directly on CrewAI, which is the framework this project depends on. The difference is where the domain knowledge lives.

If you wire it yourself, you write the agent definitions and task prompts in Python and you own the AKShare wrappers. This project puts agents in config/agents.yaml and tasks in config/tasks.yaml, so the roles, goals and backstories are editable without touching code, and the tool groups are already assembled and named. That is the entire value proposition: a pre-built A-share tool layer plus a four-role prompt structure you can tune.

The trade-off is that you inherit the author's choices. The four-role split, the task ordering, and the output format are fixed unless you rewrite the YAML, and the YAML is where the analysis quality actually lives. A generic CrewAI example gives you less domain scaffolding but no inherited assumptions about how A-share analysis should be decomposed. If you disagree with splitting sentiment from capital flow, you are editing someone else's structure rather than starting from a blank file.

A second comparison point is any single-agent setup with the same AKShare tools. One agent with all the tools is simpler and cheaper to run, and for a single stock question it may be enough. The four-agent design pays off only if you want the separate perspectives recorded separately, which is a documentation preference as much as an analysis one.

Maintenance, Upgrades and the Licence Question

The last push to the default branch was on 2026-05-07. The repository is not archived, but four months without a push means you should plan to maintain your own fork rather than wait for fixes. There are no releases, so there is no version to pin and no changelog to read before upgrading. Upgrading means pulling main and re-reading the diff yourself.

The dependency surface is the real upgrade cost. CrewAI moves quickly, and langchain imports appear in the README's own example (from langchain.llms import Ollama, from langchain.chat_models import ChatOpenAI). LangChain has reorganized those import paths across versions. If the example in the README no longer matches your installed langchain, that is your first signal that the project's pinned range and your environment have diverged, and you will be the one reconciling it.

On licensing: the README states the project uses the MIT licence. The repository metadata carries no licence identifier and no LICENSE file appears among the top-level entries. MIT is permissive and would let you use this commercially, but a README line is not the same artifact as a licence file, and this is exactly the kind of gap a legal review will flag. Treat the licence status as unconfirmed until a LICENSE file exists, and note that AKShare's own licence and the terms of the data sources it scrapes are separate questions the README does not address at all.

Editorial conclusion

Adopt it if you already run CrewAI and want a working A-share research scaffold whose agent roles and tasks live in YAML you can edit without touching Python. Do not adopt it if you need a maintained dependency, a documented licence file, or anything resembling execution or backtesting, because the repository has not been pushed since 2026-05-07, ships no releases, and its only stated licence claim is a README line. Before you trust a run, verify three things: that AKShare still returns the fields the tools expect for your stock code, that your chosen model actually follows the task output format in config/tasks.yaml, and that the four agents disagree in useful ways rather than converging on the same summary.

Frequently asked questions

Official sources

1. Issues
2. liangdabiao/easy_investment_Agent_crewai on GitHub
3. Project website
4. README