KaleidoSphere for DeepSeek Harness

Native database analysis inside DeepSeek Harness — one plugin install, no separate KaleidoSphere checkout, service, endpoint, or startup step.

Preview compatibility: DeepSeek Harness dsh-v0.1.0-rc.8 / @deepseek-ai/[email protected] at 141eb6fef83422698aef7a981029e843e8161534. DSH is a Developer Preview and breaking updates may require a plugin release.

See the evidence-backed compatibility matrix for the exact host, Node.js, operating-system, and database support boundaries.

Install and try the fixture

Install DSH and pnpm, then add the immutable plugin release to DSH's shipped one-shot agent profile:

npm install --global @deepseek-ai/[email protected] [email protected]
dsh plugin --profile headless add github:JoFe2/kaleidosphere-dsh-plugin#v0.1.0-preview.3
dsh --profile headless --dump-config
dsh --profile headless "Analyze the configured KaleidoSphere database and report its engine and snapshot digest."

The plugin default uses the bundled exact MSSQL synthetic fixture, so these six native tools are available immediately without a separate KaleidoSphere installation:

• kaleidosphere_status
• kaleidosphere_discovery
• kaleidosphere_analyze
• kaleidosphere_plan
• kaleidosphere_preview
• kaleidosphere_readback

A typical agent flow is status → analyze → discovery/plan/preview → readback. Every result is wrapped in the released KaleidoSphere External API v2 integrity envelope and a K1 evidence receipt. The intent set comes from the released K2 closed-intent contract. Install the same bundle into the active DSH profile when using another DSH surface; the profile must include an agent runner such as the shipped headless or Web app.

Short demo:

User: Analyze the configured database, propose a weekly order-value view, and show the readback.
Agent: kaleidosphere_status → kaleidosphere_analyze → kaleidosphere_plan
       → kaleidosphere_preview → kaleidosphere_readback
Result: deterministic fixture snapshot 293a896156d8f6269c4ad33e8d632da653ea180d35a4ea5f390b0be52ce3e44a

Configure a real source

KaleidoSphere v0.16.0's supported main runtime paths are Microsoft SQL Server and Oracle. The plugin includes both required client paths (including the exact Oracle Thin driver), so no separate driver approval or KS install is required. It accepts the existing closed chimpmaera.db/analyze-profile/v1 object directly; it does not invent a universal connection schema. Copy the relevant example into $DSH_HOME/profiles/headless/cordis.patch.yml and adjust it:

• examples/live-mssql.patch.yml
• examples/live-oracle.patch.yml

Put the database password in the environment variable named by adapter.passwordEnv, then start DSH. Password values are neither accepted in plugin config nor returned by tools.

export KS_MSSQL_PASSWORD='...'
dsh --profile headless "Analyze the configured KaleidoSphere database."

The configured principal must be read-only. The bundled KS query packs contain allowlisted metadata SELECT statements; raw rows, free SQL, credentials, source writes, and persistent Superset mutation are outside this plugin's tool surface. PostgreSQL remains the bounded v0.16.0 pilot and is not advertised here as a main live plugin path.

Advanced tool exposure

…