dsh-db
Database workbench: playground.db connection, table tree, SQL editor, export
A database workbench plugin for DeepSeek Harness: SQLite connection manager + SQL editor + data-grid panel + an agent-facing db tool — plus the sqlite backend that DSH's storage hub documents but does not ship.
Built on Node's built-in node:sqlite (DatabaseSync): no native dependencies, no build step, no Electron ABI roulette.
中文 | English
What you get
| Surface | Contents |
|---|
| Panel | A "Database" icon in the left sidebar → full-page workbench: connections, table tree, SQL editor, paginated/sortable/filterable data grid, inline cell editing, CSV/JSON export |
| Agent tool | One db tool with an action parameter, so the agent in a session can inspect and query your databases |
| Storage backend | Registers as backend sqlite on ctx.storage, so other plugins can route their ctx.storageDomain domains to SQLite |
Panel actions and agent tools call the same host methods. One operation, two callers.
Install
git clone https://github.com/406464/dsh-db.git
cd dsh-db && npm install && npm test # self-test, no DSH required
Then install it into DSH — dsh plugin forwards its arguments to pnpm in the profile directory:
# Windows — from npm
& "D:\deepseek harness\resources\runtime\cli\bin\dsh.cmd" plugin --profile desktop add @406464/dsh-db
# macOS desktop
DSH_CLI="/Applications/DeepSeek Harness.app/Contents/Resources/runtime/cli/bin/dsh"
"$DSH_CLI" plugin --profile desktop add @406464/dsh-db
Until the npm release lands, install straight from GitHub:
"$DSH_CLI" plugin --profile desktop add github:406464/dsh-db
Or install from a local directory via the Plugins page, or ask the agent to run install_bundle on the cloned folder.
Restart DSH afterwards: host-side modules load once at startup. If the sidebar entry is missing, check that the bundle is enabled on the Plugins page.
Configuration
Set these on the plugin row in your profile's cordis.patch.yml (the user patch layer survives upgrades):
- name: 'dsh-db'
config:
root: ~/.dsh/plugin-data/dsh-db # default: derived from the profile mount point, else ~/.dsh
defaultLimit: 200
maxLimit: 5000
maxRows: 1000
maxBytes: 262144
agentDefaultLimit: 50
allowWrite: false
allowAttach: false
allowAddConnection: false
registerStorageBackend: true
allowDirs: []
Route a domain to SQLite:
- name: '@deepseek-ai/dsh-storage'
- name: 'dsh-db'
- name: '@deepseek-ai/dsh-storage-domain'
config:
backend: json
routes:
mydomain: sqlite
Safety defaults
- Only registered connections. A path must sit inside an allowed root (plugin data dir + workspaces +
allowDirs). Unregistered files are unreachable.
- Two gates for writes.
allowWrite plus an explicit confirm: true per call.
- Statement guard. Unrecognised leading keywords are rejected;
ATTACH/DETACH off by default; load_extension always rejected.
- Bounded results. Row-count and byte caps, with
truncated: true instead of dumping a table into the model context.
- Loopback-only HTTP.
/dsh-db/api/* accepts local requests and POST only.
Errors carry stable codes: write-disabled, confirm-required, forbidden-sql, path-out-of-scope, multi-statement, not-found, sql-error.
How this differs from similar plugins
dsh-db-studio already exists on npm (SQLite / PostgreSQL / MySQL, with an in-conversation database tab, a /db command, and query history). Different tradeoffs:
| dsh-db (this project) | dsh-db-studio |
|---|
| Databases | SQLite only | SQLite + PG + MySQL |
| Dependencies | zero (Node's built-in node:sqlite) | pg + mysql2 |
| storage-hub sqlite backend | yes — other plugins can route domains to SQLite | no |
In-conversation binding / query history / /db | no | yes |
| Writes by default | off (allowWrite: false + per-call confirm) | on |
Pick theirs for company PG/MySQL and conversation-bound querying. Pick this one for a dependency-free local SQLite workbench that also fills the storage hub's sqlite slot.
Releases and contributing
npm publishing, CI, code layout and the one-time NPM_TOKEN setup live in CONTRIBUTING.md. Users don't need any of it.
Known limits (stated plainly)
DatabaseSync is synchronous and runs on the host's main thread. A full-table scan stalls the whole Harness — node:sqlite exposes no statement cancellation. Row/byte caps and explicit LIMIT are the current mitigation; moving heavy queries to a worker thread is on the roadmap.query accepts a single statement (use exec for batches).- Connection-table changes made outside the host are picked up via file mtime, but host code changes need a restart.
Development
npm run check # syntax-check every file
npm test # 27 assertion groups: classification, guards, schema, grid, writes, export, storage contract
node test/seed-demo.mjs # create a playground.db so the panel and tool have something to show
License
MIT
