Skip to content

About

An open-source AI research terminal. Research before conviction.

Resources

Stars

3 stars

Watchers

0 watching

Forks

Repository files navigation

WhyJ Quant

Research before conviction.

An open-source AI research tool for exploring markets, testing ideas, and keeping the evidence. Bring your own models and data sources.

Two product forms share the research foundation: Desktop for Windows/macOS users and CLI/TUI for Linux professionals, building on the existing terminal, command, and automation capabilities. Desktop currently ships for Windows; macOS delivery and acceptance are pending. See the product scope. The local Web workspace is a development/validation surface, not a third product line.

Website / 中文 / Documentation / Blog

WhyJ Quant website demo: research tools, sample analysis, and a market overview

Interactive website demo with synthetic data. The CLI runs in your own terminal.

Quick start

For Windows Desktop, choose the installer and verify its checksum on the GitHub Releases page. For the CLI, install the published package, then launch the terminal:

npm install -g whyj-quant
whyj

If a Windows npm upgrade is interrupted and whyj is unavailable, see CLI installation recovery before retrying.

Open /config to connect your model and market-data providers. Use /doctor to check the setup, then ask a question:

Analyze the returns and risk of 000001.SZ from 2025-01-06 to 2025-06-30.
Show the actual data source, coverage dates, assumptions, and missing inputs.

To save reports, enable Auto Artifacts in /config before submitting a research request. /artifact opens existing reports; /resume returns to a saved session.

Provider credentials and optional Python dependencies depend on your data sources. Model and data services may charge for usage. See the configuration guide and provider reference.

Run from source

Development requires Node.js 22.19+ and Bun 1.3.14 (the build tool and test runner). From the repository root:

bun install --frozen-lockfile
npm run dev

The checkout and npm release may differ. Check whyj --version and the changelog for the version you are using.

Source checkouts also include an experimental local Web research workspace: bun run workspace:dev, then open http://127.0.0.1:4175/. It connects the configured Agent to one explicit-stock daily-bar task, with citations, manual saving and independent review. Requires Python with AKShare and pandas; see current persistence and recovery boundaries. This entrypoint is not part of the published CLI package or the website demo.

Features

  • Research in context. Connect market data and quantitative tools in one conversation.
  • Test an idea. Explore factors, backtests, and risk with explicit dates and data coverage.
  • Keep the evidence. Review reports and tool calls, then resume the source research.
  • Work your way. Choose models, data providers, skills, and workspace preferences.

Commands

Command Purpose
/config Connect models and data; adjust the workspace
/doctor Check the research environment
/portfolio Manage observation groups
/skill Choose, install, remove, and run skills
/mcp Choose and manage MCP presets for Pi
/artifact Browse reports and restore their source research
/resume Continue a saved session

Use /help in the terminal or whyj --help in your shell. Command help provides detailed parameters and one-shot usage; the command catalog defines the public entries.

Documentation

Most technical references are in Chinese. The website supports English and Chinese; articles will appear on the Blog.

Development

The package exposes npm script entrypoints. The built terminal and one-shot CLI run on Node. The #76 source branch adds whyj web [--no-open] (also with --json): this mode starts or reuses a local Host using the CLI's Node or Bun runtime. It is not yet a published release. See the CLI entry for build prerequisites and the remaining TUI integration.

npm run dev            # Source CLI with watch mode
npm run build          # Build the Node CLI into dist/
npm start             # Run the built CLI (build first)
npm test              # Test fence, module tests, and repository integration tests
npm run check         # Types, architecture, tests, CLI build, and documentation
npm script Scope
npm run typecheck Source, module-local development entrypoints, and root tests
npm run check:architecture Production dependency boundaries; excludes manual dev consumers
npm run test:modules Tests owned by individual src modules
npm run test:integration Cross-module checks in tests/architecture/ and tests/app/
npm run test:watch Watch both test roots
npm run docs:build / npm run check:docs Offline blog generation and documentation checks

Maintenance tools live alongside their owners under src/<module>/dev/; there is no shared root scripts entrypoint. Production code cannot import dev tools or test fixtures. Module tests stay inside their module, while repository architecture and application integration tests live in root tests/.

Explicit model checks consume provider usage; --help does not:

npm run bench:model -- --help
npm run eval:research -- --help

On Windows PowerShell, if the npm.ps1 shim consumes forwarded flags, use npm.cmd run bench:model -- --help (and the same npm.cmd spelling for other forwarded commands).

The product website lives in site/ and has its own build:

npm --prefix site run dev
npm --prefix site run build

License

MIT. Built for thoughtful research.

Research software, not investment advice. Observation groups are not brokerage accounts. Settings, sessions, and reports are stored locally; analysis sends necessary requests to your chosen providers. Review reports before sharing them.

About

An open-source AI research terminal. Research before conviction.

Resources

Stars

3 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages