KabuForge
Menu
GitHub日本語

Get started

Start with AI-assisted installation, then prepare data, run your first study and inspect results.

Let an AI assistant help you set up

Ask Codex, Claude or Cursor with local file and terminal access to help install and configure KabuForge. Start by having the assistant obtain the source, create a Python virtual environment and check the installation. This setup assistance does not require MCP. In a chat-only environment, run the suggested steps yourself.

Choose stable or the unpublished main candidate before starting. Replace the installation and research workspace paths in the prompt below.

Copy a prompt into your AI assistant

Press Copy below and paste the prompt into an assistant with local file and terminal access. It will work through installation, MCP connection and supported personalization within the permissions you approve. Client permissions and manual steps may require your input.

Help me install KabuForge, connect it through MCP, and personalize its setup for my research.
Official repository: https://github.com/Siyuan-chat/KabuForge
Get started: https://kabuforge.com/en/get-started/

1. Check my OS, Python and current AI assistant. Ask me to confirm installation and research workspace paths and whether to use stable v0.1.1 or the unpublished main candidate. The latest GUI tutorials apply to main.
2. Before execution, explain what will be installed, which files will change, and the terminal, network and directory permissions needed. Obtain my approval and stay within that scope.
3. Follow the selected version's README and installation guide. Install in a dedicated virtual environment without overwriting existing environments. Verify with kabuforge doctor and the official synthetic offline demo.
4. Read the official MCP guide for my assistant and configure absolute executable and workspace paths. Before writing client configuration or granting tool access, show the changes and permissions and obtain my approval. Preserve existing settings. Explain any restart or manual steps I must perform. Verify tool listing and read-only access.
5. Briefly ask about my experience, research goals, available data, symbols and date range, research frequency, and GUI/report preferences. Suggest supported settings and a first tutorial. Show proposed changes and obtain approval before applying them. Do not claim unsupported features are configured.
6. Obtain separate scoped approval for private-data access, external transmission, downloads, paid services and research execution. Do not connect brokerage accounts or trade. Never ask for passwords or API keys in chat.
7. Report the selected version, paths, MCP result, applied settings, checks actually completed, unresolved issues and next steps. Clearly identify anything you could not execute.

Download the Chinese setup prompt

Verify setup, then connect research tools

After setup, check the selected version, virtual environment path, diagnostic results and demo output. Ask the assistant to identify checks it did not run and explain any errors and next steps.

To call KabuForge research tools from an assistant, configure MCP next. MCP connects the assistant to the research tools. Use your client guide, configure absolute executable and workspace paths, then verify that tools can be listed.

Inspect available KabuForge tools and registered inputs read-only. Explain what is needed for my first offline study. Do not run research or modify data yet.

Once ready, continue with the five steps below.

Five steps

For stable v0.1.1, start with the synthetic offline quickstart. The GUI courses below apply to the unpublished main source candidate. Install it separately from your stable environment.

  1. Install the main source candidate in a separate virtual environment and open that environment’s GUI.
  2. Select, validate and freeze your own local input.
  3. Choose a strategy such as Price momentum and check the frozen input and settings.
  4. Review preflight, then run the backtest.
  5. Inspect results, input SHA and run history.

Screens and inputs

Screenshot, receipt and log index

Market data is not bundled. To begin with synthetic inputs, use the offline demo.

Applies to: main source candidate (unpublished) · Public source docs/en_US/GUI_RESEARCH_COURSES.md

Reviewed · KabuForge