dsh-chatbi

Readme

# dsh-chatbi

DeepSeek Harness (DSH) plugin for **ChatBI-style tooling**: database connections, MCP/API/SQL tool registry, and agent tools — **fully self-contained**, no external ChatBI Java backend required.

Not affiliated with DeepSeek.

## Architecture

All data lives in a local JSON store on the Host:

```
$DSH_HOME/chatbi/store.json
```

| Area                 | Storage     | Execution                                                                   |
| -------------------- | ----------- | --------------------------------------------------------------------------- |
| Database connections | local store | **SQLite** via Node built-in `node:sqlite`; other types stored for metadata |
| MCP servers & tools  | local store | HTTP POST to configured server `baseUrl`                                    |
| SQL tools            | local store | Template → `chatbi_sql_execute`                                             |
| API tools            | local store | HTTP fetch to configured `baseUrl` + path                                   |

Settings UI layout follows chatbi-ui tab patterns, built with **Tailwind CSS v4**.
Compiled utilities are embedded into `lib/client.js` and injected at runtime
(DSH ModuleLoader does not load sidecar `.css` files).

## Install

```bash
dsh plugin --profile web add ./
```

Node ≥22.

## Settings

**Settings → ChatBI**:

- Optional **data directory** override (default `$DSH_HOME/chatbi`)
- Tabs: 概览 / 数据库 / MCP 工具 / API 工具 / SQL 工具

## Agent tools

| Tool                 | Purpose                    |
| -------------------- | -------------------------- |
| `chatbi_status`      | Local store summary        |
| `chatbi_sql_execute` | Run SQL (sqlite supported) |
| `chatbi_mcp_call`    | Invoke registered MCP tool |
| `chatbi_api_call`    | Invoke registered API tool |
| `chatbi_list_*`      | List MCP / SQL / API tools |

Skill `dsh-chatbi` documents the workflow.

## Verify

```bash
npm install --legacy-peer-deps
npm test
npm run build
```

## Limitations

- MySQL/PostgreSQL: connections are stored locally; live SQL needs drivers not bundled (use sqlite or MCP/API tools).
- MCP invoke assumes an HTTP endpoint on the configured server (generic JSON body).
- Seed MCP/SQL/API tools by editing `store.json` or extend the UI later.

## License

MIT — see [LICENSE](LICENSE).