DeepSeek Harness Plugin Hub

Publish and manage complete Harness Profiles. Discover Plugins for your next setup.

Explore

PluginsPresetsDocsNews

Community

Publish a pluginContactReport an issue

Resources

Plugin Hub on GitHubDeepSeek HarnessSystem statusPrivacy notice
© 2026 DeepSeek Harness Plugin HubPowered byPaxTech

Independent and unofficial. Not affiliated with, authorized by, or endorsed by DeepSeek.

Database — DSH Plugin for DeepSeek Harness
DeepSeek Harness Plugin Hub
ProfilesPluginsCategoriesNewsDocsSign inManage Profiles
ProfilesPluginsCategoriesNewsDocsSign in
← Plugins
D

dsh-database

Database

MySQL, Oracle, Redis and Kafka inside DeepSeek Harness. Written entirely with AI.

The plugin will be installed here. Keep web if you are unsure.

npx -y @deepseek-ai/dsh plugin --profile web add github:zhuoxiaoshuai/dsh-database#5d2b900b0317200b1866275c7cf48814fe84b52d
READMECompatibilityVersions
WorkbenchMySQL resultsOracle catalogRedis keysKafka topic

Compatibility and provenance

Database is published as dsh-database and currently resolves to version 0.1.0-alpha.12.18. The Hub verifies its manifest and preserves the exact installation source for reproducible installs.

DSH compatibility
*
Runtime surfaces
web
Release source
github
Registry updated
10/9/2026

Versions

0.1.0-alpha.12.18prerelease
10/9/2026
Show 1 more versionCollapse versions
0.1.0-alpha.12.15prerelease
9/30/2026

Related plugins

Loading related plugins…

Latest
0.1.0-alpha.12.18
DSH
*
HMR
Process restart
Tree shaking
Safe tree shaking not declared
Unpacked size
Unavailable
Files
Unavailable
Surface
web
License
MIT
Source
github
GitHub
★ 0
Weekly downloads
0
Last push
10/9/2026
View source ↗
README badge

Click the badge to copy Markdown for your README.

Do you maintain this Plugin?Claim benefit · Priority security scan

Verify the GitHub repository declared in package.json to manage this listing. After you claim it, Hub will prioritize a security scan of the current version and publish the result when it passes.

Claim this Plugin →
Report an issue

Related plugins

More verified plugins in integrations-communication.

Acp App@deepseek-ai/dsh-acp-appThe dsh ACP profile bundle: automation-only JSON-RPC stdio and process lifecycle over dsh-baseAb Publishdsh-ab-publishBrowser-automation publishing tool that logs into and posts to arbitrary websites, in one publishable plugin packageAb Wechat Scrapedsh-ab-wechat-scrape--- description: "The out-of-tree wechat_scrape tool: retrieves WeChat official-account articles, filters promotional blocks, and saves each as a Markdown file, for maintainers of this machine's dsh profiles." kind: "package-reference" ---Remote Web Ui@linxin666/dsh-remote-web-uiScan-to-pair remote access for the dsh web GUI that shares one official interface: a QR beside the settings button pairs phones and PCs into the same Web GUI (a portrait-touch adaptation layer for phones, full desktop on PCs) through one-time tokens and r

README

DSH Database

Connect to MySQL, Oracle, Redis and Kafka from the right sidebar of a DeepSeek Harness conversation. Browse objects, run queries, and share a query editor with AI. SQL grid maintenance and a knowledge library support database work; bounded Kafka reads and SIT repair actions support diagnosis.

This repository was written entirely with AI.

中文 · Install · First use · Permissions · Development · Docs

Workbench

Names in the screenshots are test data.

Features

SourceCommon tasks
MySQL / OracleDatabase/schema, table, column and structure browsing; query/object tabs, SQL batches, execution plans, grid maintenance and SQL knowledge analysis
RedisDB/key browsing, types, TTL, values and a single-command console
KafkaTopics/groups, partitions, watermarks, lag, message previews, bounded search and SIT publish/repair actions

All sources share query, AI Query, history and knowledge entry points. Editors and results retain source-specific semantics.

Install

dsh plugin --profile web add https://github.com/zhuoxiaoshuai/dsh-database/releases/download/v0.1.0-alpha.12.15/dsh-database.tgz

Restart dsh web afterwards. Node.js 24 or newer is required. The git repository does not ship lib/, so do not install from github:.

From a clone you can also pack it locally:

npm ci --legacy-peer-deps
npm pack
dsh plugin --profile web add ./dsh-database-0.1.0-alpha.12.18.tgz

To uninstall:

dsh plugin --profile web remove dsh-database

Do not copy cordis.patch.yml into the profile by hand.

On Desktop, dsh is not on PATH. Open DSH Terminal from the tray and run:

dsh plugin --profile desktop add https://github.com/zhuoxiaoshuai/dsh-database/releases/download/v0.1.0-alpha.12.15/dsh-database.tgz

It is the same plugin on a different profile. --profile web does not install into the desktop app.

The download URL above targets the alpha.12.15 release package. The source currently declares alpha.12.18; use the filename printed by npm pack, which builds automatically. Check acceptance evidence separately for the release package and local source.

First use

  1. Install, restart the chosen profile, and open the Database tab in the conversation's right sidebar.
  2. Click + beside 我的连接, select a source, and enter its address, credentials and database/Service Name/DB or brokers. Choose SIT, UAT or PVT.
  3. Optionally click 测试连接 to check the settings, then 连接. Testing alone does not add a saved connection.
  4. Select a database/schema for SQL. Run SELECT 1; for MySQL, SELECT 1 FROM DUAL; for Oracle, PING for Redis, or TOPICS for Kafka in the query page.
  5. For AI collaboration, open AI Query and describe the connection, target and task in the conversation, such as “Read the current SQL and explain what it does.” Check the target and permissions before execution.

AI collaboration and permissions

Typing in AI Query or clicking 接管 takes control of the document. AI cannot overwrite the text while you control it. Finish editing and click 归还 AI to let the model continue. The ordinary query page and AI Query are separate entry points.

OperationSITUAT / PVT
Human SQL, ordinary or taken-over coeditReads and DML; auto-commit within account privilegesReads and DML; auto-commit within account privileges
AI SQLReads and DMLRead-only
Grid edits / structured DDLHuman preview and confirmationDisabled
Human Redis consoleSubject to ACL and command validationSubject to ACL and command validation
AI RedisStructured reads and redis_executeStructured reads; redis_execute disabled
Kafka readsAllowedAllowed
Kafka writesHuman confirmation or AI tools; broker permissions still applyDisabled

UAT/PVT do not prevent human SQL or human Redis writes. Use a read-only database account or Redis ACL to prohibit writes. AI SQL results on SIT return unredacted cells to the model; ordinary human queries do not automatically send their results to it.

Imported dev/test labels map to SIT, staging to UAT, prod to PVT, and unknown labels to UAT. Passwords stay in Host memory by default. On Windows, optional 记住密码 encrypts them using current-user DPAPI. See SECURITY.md.

An interrupted write or a lost acknowledgement may be reported as unknown: the change may already exist. Check the actual target before deciding what to do next. Unknown writes are never automatically replayed. Reconnecting invalidates old requests and execution targets.

MySQL

MySQL results

The driver is mysql2, default port 3306. The tree is database → table → column. Identifiers use backticks; paging uses LIMIT / OFFSET. mysql, information_schema, performance_schema, and sys stay in the tree.

SELECT 9007199254740993 AS id;

When the account can read them, the catalog shows columns, indexes, constraints, and SHOW CREATE. Otherwise it explains why, instead of failing silently.

Filter, sort, and paging run on the server. The default is 100 rows; the cap is 500 rows or 1 MiB. Cancelling a query does not drop the login.

Grid edits use parameterized statements: preview, confirm once, and match the original primary-key row. A conflict rolls back and keeps the draft.

DDL allows at most 20 steps. Approval lasts 5 minutes. Destructive steps ask you to type the table name. The run stops on the first error; there is no all-or-nothing DDL rollback.

database_execute_sql runs up to 8 statements, 100 rows each. It is read-only on UAT/PVT and permits DML on SIT. The batch is checked before dispatch, runs in order, and stops at the first error. Earlier commits are not rolled back when a later statement fails.

BIGINT is returned as a string. Binary cells show readable UTF-8 text or hexadecimal bytes (BIT uses binary digits); binary columns stay read-only in the grid.

Oracle

Oracle catalog

The SQL page is the same as MySQL. Objects are organized by schema. Unquoted SQL identifiers are folded to uppercase; double-quoted identifiers preserve case. If a schema named after the username exists, that is the default. Identifiers are quoted; paging uses OFFSET … ROWS FETCH FIRST … ROWS ONLY; plans use EXPLAIN PLAN FOR. q-quote works; # comments do not.

The driver is oracledb Thin, default port 1521. The form supports Service Name and SID; live SID acceptance remains NOT_RUN. NUMBER and timestamps are fetched as STRING. Views and synonyms expose metadata only. A query that includes a LOB column fails.

Redis

Redis keys

Redis 7.2 or newer is required. The tree is DB → keys. SCAN paging can return empty pages or duplicates; the UI says when the cursor is unfinished.

PING

Standalone, one sentinel, or one cluster seed are supported. ACL, TLS, and a custom CA are optional. Sentinel takes a single address; the same user and password are used for sentinel and Redis. In cluster mode the host is a seed and DB must be 0.

The key browser and the command console use different connections. The console runs one CLI-quoted command and then closes. SELECT or MULTI in the console does not change the tree.

The human command console can execute commands on SIT, UAT and PVT. AI redis_execute is SIT-only; other environments use structured read tools. DSH_REDIS_COMMAND_BLACKLIST is empty unless you set it; Redis ACL still applies.

Kafka

Kafka topic

Search and page topics/groups, inspect partitions, lag and settings, locate offsets by time, and search bounded message ranges. PEEK uses a temporary group and does not commit business consumer offsets.

TOPICS SEARCH "orders"
GROUP "billing" TOPIC "orders"
PEEK "orders" PARTITION 0 FROM LATEST LIMIT 20

Replace the topic and group with actual names. SIT permits single/limited batch publishing, replaying complete messages, compacted-topic tombstones, dsh-test- topic creation and offset changes for groups without running members. Existing messages cannot be modified in place.

Authentication supports none, TLS with CA, PLAIN and SCRAM-SHA-256/512. Kerberos, OAuth and client certificates are not supported. See the Kafka guide for full commands, byte/result limits, confirmation and unknown outcomes.

AI tools

database_status without a topic returns the live connections and the next call's connectionId and generation. A topic loads one usage section. A missing connection or a stale generation returns the current connection list. When that source has exactly one live connection, connectionId and generation can be omitted.

kafka_scan and kafka_set_group_offsets take spec as an object, kafka_produce takes headers as an object, and kafka_produce_batch takes messages as an array. kafka_peek from is BEGINNING, LATEST, or OFFSET.

Expand tool reference
ToolNotes
database_statusLive connections and the next call's arguments. A topic returns only that usage section
database_catalogschemas / tables / table. Do not query information_schema
database_execute_sqlaction=read returns the current AI Query text. With sql, it executes; read-only on UAT/PVT, writes allowed on SIT
database_templatesSave and search text. Does not execute
database_read_collabOpen query tabs
database_import_connectionsRegister hosts. No password
redis_status redis_keys redis_valueSCAN, type, TTL, value
redis_executeSIT only, one command
kafka_status kafka_topics kafka_describeTopics, partitions, watermarks
kafka_peekPeek one partition. from is BEGINNING, LATEST, or OFFSET. Refused after editor takeover
kafka_groups kafka_groupSearch/page groups; inspect state and members
kafka_group_topics kafka_group_topicPage associated topics; inspect committed positions and Lag
kafka_topic_config kafka_time_offsets kafka_scanTopic settings, time-based offsets, bounded cross-partition search. kafka_scan spec is an object
kafka_produce kafka_produce_batch kafka_tombstoneSIT single/batch publish and compacted-topic tombstone. headers is an object; messages is an array
kafka_create_topic

Troubleshooting

ProblemWhat to do
Plugin installed but no tabRestart the profile you installed into, then refresh; avoid duplicate manual cordis.patch.yml entries
dsh missing on DesktopOpen DSH Terminal from the tray
/plugins/database/connections clashCheck whether an older remote-exec still provides the database UI
AI cannot update or run current textCheck document control; finish saving and return control to AI. Refresh connection status after reconnecting
Oracle query includes a LOBSelect ordinary columns; on-demand LOB viewing is not connected yet
Redis console SELECT leaves the tree unchangedChoose the DB in the tree; each console command uses a separate connection
Write result is unknownRead the target to verify the change before resubmitting
Query times outSQL driver query timeout is 25 seconds; Host query deadline is 32 seconds. Maintenance and source operations have separate limits; see timeout configuration

Development

Use Node.js 24 or newer. Install dependencies and check from the repository directory:

npm ci --legacy-peer-deps
npm run check
npm run test:package-closure

check runs Host/client type checking, unit tests and builds. Package closure separately checks shipped runtime files. Run browser and live-source acceptance separately:

CommandRequirements and scope
npm run test:sql-workspaceMicrosoft Edge; controlled bridges for MySQL/Oracle client acceptance
npm run test:workspace-racesMicrosoft Edge; coedit save/control/target races
npm run test:execution-boundaryMicrosoft Edge; controlled execution-boundary and grid-value checks
npm run test:mysql / test:oracleDocker Linux Engine; owned temporary databases with cleanup, not Oracle 19c/SID acceptance
npm run test:redisDocker and OpenSSL; temporary plain/TLS Redis acceptance. Set DSH_OPENSSL if needed
npm run test:kafkaDocker and OpenSSL; isolated broker acceptance
npm run test:hostHarness installation directory (DSH_DESKTOP_APP) and Chrome (or a supported browser via DSH_TEST_BROWSER); installs and starts a Web host in a disposable profile

Windows PowerShell example:

$env:DSH_DESKTOP_APP = 'C:\path\to\DeepSeek Harness'
npm run test:host
# Optional live-source checks; database fixtures require Docker
$env:DSH_TEST_DATABASES = '1'
$env:DSH_TEST_REDIS = '1'
$env:DSH_TEST_KAFKA = '1'
npm run test:host

In a POSIX shell, use DSH_DESKTOP_APP="/path/to/install" npm run test:host. Reports are written under artifacts/. Controlled bridges, isolated live sources and installed-host acceptance are recorded separately. test:host does not establish live Desktop UI or model-call acceptance.

Compatibility and acceptance records

Current drivers: mysql2 3.24.4, oracledb 7.0.1, redis 6.2.1 and kafkajs 2.2.4. Historical validation includes Harness 0.1.2-rc.1, 0.1.7-rc.2 and 0.2.0-rc.2, MySQL 8.4/8.0.32, Oracle Free 23/19c Service, Redis 8.10.2 and test Kafka instances. These records do not automatically cover later source changes or every authentication combination.

RecordCoverage and limits
Execution-boundary repair (2026-10-05)Records passing type checks, 768 unit tests, controlled browser checks, build and package closure. Final four-source live reruns, Oracle 19c/SID, installed Desktop and real model calls remain NOT_RUN
Kafka repair extension (2026-10-04–05)Controlled regressions and stage-specific isolated broker runs; final-code live rerun and installed/model acceptance remain NOT_RUN
Historical installation/source acceptanceConsult for the versions, environments and scope tested at that time

These are dated records. This README update did not rerun database, Desktop or model acceptance. Establish the validation status of current source using its build and actual reports.

Documentation and feedback

  • Documentation index, architecture, source onboarding
  • Kafka guide, implementation records
  • Issues, security

License

MIT

kafka_set_group_offsets
SIT test topic creation and inactive-group offset adjustment