database-dialects

name

database-dialects

description

Use when the user mentions a specific database product, asks about supported data source types, picks a connection kind, or asks why a feature (ER / diagnostics / mutation) is unavailable on a given dialect. Triggers on MySQL / PostgreSQL / Oracle / SQL Server / SQLite / DuckDB / ClickHouse / TiDB / MariaDB / OceanBase / StarRocks / Apache Doris / Trino / Presto / Dameng / Apache Hive / GaussDB / 方言 / dialect / 数据库类型 / 数据源类型 / 连接类型. Covers per-dialect connection kind, default port, JDBC driver, schema visibility, SQL splitter family, risk-tier highlights, and ER / diagnostics support status.

Database Dialects Skill

When to use

• The user asks "DataTalk supports which databases?", "支持哪些数据库?", or "can I connect to ?"
• The user picks or names a connection kind (mysql, tidb, clickhouse, oceanbase, gaussdb, ...) and you need its port / driver / schema model
• A tool returns dialect_unsupported and you need to explain to the user which features are off on that dialect
• The user is composing SQL and asks about per-dialect risk classification (L1 / L2 / L3) or splitter behavior (DELIMITER, GO, dollar-quoting)
• The user types Chinese / English aliases ("达梦", "Dameng", "PingCAP TiDB", "蚂蚁 OceanBase") and you need the canonical kind

When NOT to use

• Generic SELECT execution or aggregation workflow — use [[sql-execution]]
• Connection creation, switching, use xxx, confirm=true / confirmationToken mutation protocol — use [[connection-management]]
• SQL execution errors, table doesn't exist probe path, multi-candidate disambiguation — use [[sql-error-diagnostics]]
• ER Inspector / Designer protocol details — use [[er-tabs]]
• Choosing a chart type for query results — use [[charts-and-dashboards]]

Common dialects matrix

Dialect	Kind	Default port	Driver	ER (Inspector / Designer)	Diagnostics (Day-1)
MySQL (baseline)	mysql	3306	MySQL Connector/J	supported / supported	supported (EXPLAIN, lock, pool, space)
PostgreSQL (baseline)	postgres	5432	PostgreSQL JDBC	supported / supported	supported
H2 (baseline)	h2	embedded	H2 JDBC	dialect_unsupported / dialect_unsupported	dialect_unsupported
SQLite (baseline)	sqlite	embedded	SQLite JDBC	dialect_unsupported / dialect_unsupported	dialect_unsupported
MariaDB	mariadb	3306	MariaDB Connector/J	supported / supported	supported (reuses MySQL EXPLAIN)
TiDB	tidb	4000	MySQL Connector/J (wire-compat)	dialect_unsupported / dialect_unsupported	dialect_unsupported
Oracle	oracle	1521	Oracle JDBC	(Inspector dialect-supported) / dialect_unsupported	structured plan unsupported
SQL Server	sqlserver	1433	Microsoft JDBC	(Inspector dialect-supported) / dialect_unsupported	structured plan unsupported
DuckDB	duckdb	embedded	DuckDB JDBC	dialect_unsupported / dialect_unsupported	dialect_unsupported
ClickHouse	clickhouse	8123 (HTTP)	ClickHouse JDBC	dialect_unsupported / dialect_unsupported	dialect_unsupported
Apache Doris	apache_doris	9030	MySQL Connector/J	dialect_unsupported / dialect_unsupported	dialect_unsupported
OceanBase (MySQL-mode)	oceanbase	2881	com.oceanbase:oceanbase-client	dialect_unsupported / dialect_unsupported	dialect_unsupported (all 9 hooks)
StarRocks	starrocks	9030	StarRocks JDBC	dialect_unsupported / dialect_unsupported	dialect_unsupported
Trino	trino	8080	Trino JDBC	dialect_unsupported / dialect_unsupported	dialect_unsupported
Presto	presto	8080	Presto JDBC	dialect_unsupported / dialect_unsupported	dialect_unsupported
Dameng	dameng	5236	com.dameng:DmJdbcDriverX:8.1.x	dialect_unsupported / dialect_unsupported	dialect_unsupported (all 7 hooks)
Apache Hive	hive	10000 (binary)	Hive JDBC	dialect_unsupported / dialect_unsupported	dialect_unsupported
GaussDB	gaussdb	8000	com.huaweicloud:gaussdbjdbc:v2.0-8.218.0	dialect_unsupported / dialect_unsupported	dialect_unsupported

MySQL & PostgreSQL baseline (no dedicated ## section below): MySQL uses kind=mysql, default port 3306, schema selector hidden (database-only). PostgreSQL uses kind=postgres, default port 5432, both database and schema selectors visible. Both are the reference dialects against which compatibility-mode kinds (MariaDB, TiDB, Doris, OceanBase, StarRocks, GaussDB) are described. H2 / SQLite are embedded baselines used mainly in tests; their ER and diagnostics support is intentionally minimal.

MariaDB

• Connection kind: mariadb. MySQL-compatible; uses MariaDB Connector/J driver.
• Default port: 3306. Same fields as MySQL (host, port, database, username, password).
• Metadata and SQL execution follow the same paths as MySQL.
• Diagnostics reuse MySQL EXPLAIN.
• ER Designer: supported. Reuses MySQL DDL generation. kind=mariadb connections bind to mariadb dialect designers; kind=mysql connections also bind to mariadb designers. ER details — see [[er-tabs]].
• Schema visibility: database selector visible, schema hidden (same as MySQL).
• Chat-path execution: SELECT / INSERT / UPDATE / CREATE / ALTER…ADD run directly via datatalk_execute_sql; DELETE triggers the in-chat confirmationId flow; destructive DDL (DROP / TRUNCATE / ALTER…DROP / GRANT / REVOKE / DENY / KILL / SHUTDOWN / PURGE / SET GLOBAL / INSERT OVERWRITE) returns redirect_to_editor — AI opens a query_editor tab for manual execution. See [[sql-execution]].

TiDB

• Canonical kind: tidb. Reject any user attempt to map TiDB to mysql.
• Default port: 4000.
• Protocol: MySQL 5.7 / 8.0 wire-compatible. SQL splitting, formatting, and most DML/DDL behave like MySQL. The TiDB-only L3 statements listed below have cluster-wide impact and should be surfaced to the user before running them via datatalk_execute_sql.
• TiDB-only L3 (must surface confirmation): ADMIN CANCEL/PAUSE/RESUME DDL JOBS, BACKUP DATABASE, RESTORE DATABASE, IMPORT INTO, LOAD DATA INFILE, FLASHBACK CLUSTER/DATABASE/TABLE, ALTER/CREATE/DROP PLACEMENT POLICY, KILL TIDB, SET GLOBAL, BATCH ON ... INSERT/UPDATE/DELETE.
• TiDB-only L2: SPLIT TABLE ... BETWEEN ... AND ..., RECOVER TABLE, ANALYZE TABLE, ALTER TABLE ... COMPACT, ADMIN CHECK TABLE/INDEX.
• TiDB-only L1 read-only introspection (safe to suggest to the user): SHOW PLACEMENT, SHOW PLACEMENT FOR ..., SHOW PLACEMENT LABELS, SHOW TABLE <t> REGIONS, SHOW SPLIT REGIONS, SHOW STATS_HEALTHY, SHOW STATS_HISTOGRAMS, SHOW STATS_META, SHOW STATS_BUCKETS, ADMIN SHOW DDL, ADMIN SHOW DDL JOBS. Use these when the user asks about cluster topology, region distribution, statistics health, or DDL job state.
• Diagnostics (lock / pool / table_space / EXPLAIN-real / index-hints / terminate / optimize) are dialect_unsupported on TiDB Day-1. Suggest the user run EXPLAIN ANALYZE or the L1 introspection statements above manually in the Query Editor when execution plans, region distribution, or statistics are needed.
• ER Inspector and Designer are dialect_unsupported on TiDB Day-1 — see [[er-tabs]].
• User may type "TiDB", "tidb", "PingCAP TiDB". Map to canonical tidb only. Do not invent aliases.
• Chat-path execution: SELECT / INSERT / UPDATE / CREATE / ALTER…ADD run directly via datatalk_execute_sql; DELETE triggers the in-chat confirmationId flow; destructive DDL (DROP / TRUNCATE / ALTER…DROP / GRANT / REVOKE / KILL / SHUTDOWN / SET GLOBAL etc.) returns redirect_to_editor (AI opens a query_editor for manual execution). The TiDB-only L3 statements not caught by the keyword gate (e.g. BACKUP/RESTORE DATABASE, IMPORT INTO, LOAD DATA INFILE, FLASHBACK, ALTER PLACEMENT POLICY) execute directly — explicitly preview them to the user before running. See [[sql-execution]].

Oracle

• Connection kind: oracle. Has oracleServiceType field: "service" (default) or "sid".
• Default port: 1521. Fields: host, port, database (service name or SID), username, password.
• Schema context uses Oracle owner / schema. No independent catalog — service name is set at connection level.
• PL/SQL blocks risk: anonymous blocks (BEGIN...END) are high-risk. EXECUTE IMMEDIATE and dynamic SQL are flagged.
• ER Designer: unsupported. Use query_editor + read_schema instead — see [[er-tabs]].
• Diagnostics: structured execution plan unsupported. Use raw EXPLAIN PLAN FOR + DBMS_XPLAN — error-side reasoning lives in [[sql-error-diagnostics]].
• Schema visibility: schema / owner visible, no independent catalog selector.
• Chat-path execution: SELECT / INSERT / UPDATE / CREATE / ALTER…ADD run directly via datatalk_execute_sql; DELETE triggers the in-chat confirmationId flow; destructive DDL (DROP / TRUNCATE / ALTER…DROP / GRANT / REVOKE / DENY / KILL / SHUTDOWN / PURGE / SET GLOBAL / INSERT OVERWRITE) returns redirect_to_editor — AI opens a query_editor tab for manual execution. See [[sql-execution]].

SQL Server

• Connection kind: sqlserver (alias mssql is normalized to sqlserver).
• Default port: 1433. Fields: host, port, database, username, password.
• Extra connection options: sqlserverEncrypt (default true), sqlserverTrustServerCertificate (default true), sqlserverInstanceName (optional).
• Schema context uses databaseName + schema (similar to PostgreSQL).
• Risk keywords: EXEC, EXECUTE, BACKUP, DBCC, KILL, SHUTDOWN are flagged.
• GO batch splitter is not in Day-1 scope; multi-statement batches use semicolons.
• ER Designer: unsupported. Use query_editor + read_schema instead — see [[er-tabs]].
• Diagnostics: structured execution plan unsupported. Use SET SHOWPLAN_TEXT ON or SSMS — error-side reasoning lives in [[sql-error-diagnostics]].
• Schema visibility: both database and schema visible (like PostgreSQL).
• Chat-path execution: SELECT / INSERT / UPDATE / CREATE / ALTER…ADD run directly via datatalk_execute_sql; DELETE triggers the in-chat confirmationId flow; destructive DDL (DROP / TRUNCATE / ALTER…DROP / GRANT / REVOKE / DENY / KILL / SHUTDOWN / PURGE / SET GLOBAL / INSERT OVERWRITE) returns redirect_to_editor — AI opens a query_editor tab for manual execution. See [[sql-execution]].

DuckDB

• Connection kind: duckdb. Embedded analytical SQL engine — no host / port.
• Fields: databaseName (file path or :memory:), readOnly (boolean, defaults to false). No host, port, username, or password.
• Chat-path execution: SELECT / INSERT / UPDATE / CREATE / ALTER…ADD run directly via datatalk_execute_sql; DELETE triggers the in-chat confirmationId flow; destructive DDL (DROP / TRUNCATE / ALTER…DROP / GRANT / REVOKE / DENY / KILL / SHUTDOWN / PURGE / SET GLOBAL / INSERT OVERWRITE) returns redirect_to_editor — AI opens a query_editor tab for manual execution. See [[sql-execution]].
• File operations (COPY, EXPORT DATABASE, IMPORT DATABASE), extension commands (INSTALL, LOAD, CREATE SECRET), and external file / network access (read_csv, read_parquet, read_json, glob, httpfs, S3 paths) are not supported by the DuckDB embedded engine in DataTalk's deployment.
• ATTACH and DETACH are not supported.
• DuckDB file paths are backend-local only.
• Read-only connections cannot execute mutations.
• Schema context: schema selector is visible (normally main). No independent database selector.
• SQL splitter: generic (no DELIMITER, no PL/SQL, no GO).
• Diagnostics: structured unsupported. EXPLAIN, lock info, pool status, table space, index hints, terminate session, and optimize table are all unsupported.
• ER Inspector: Day-1 dialect_unsupported (embedded engine, foreign-key metadata not yet verified) — see [[er-tabs]].
• ER Designer: Day-1 dialect_unsupported.

ClickHouse

• Connection kind: clickhouse. Analytical column-store database over HTTP protocol.
• Fields: host (hostname or IP), port (default 8123), username, password, databaseName. Protocol defaults to HTTP; set port to 8443 for HTTPS.
• Chat-path execution: SELECT / INSERT / UPDATE / CREATE / ALTER…ADD run directly via datatalk_execute_sql; DELETE triggers the in-chat confirmationId flow; destructive DDL (DROP / TRUNCATE / ALTER…DROP / GRANT / REVOKE / DENY / KILL / SHUTDOWN / PURGE / SET GLOBAL / INSERT OVERWRITE) returns redirect_to_editor — AI opens a query_editor tab for manual execution. See [[sql-execution]].
• ClickHouse mutations (INSERT, ALTER, DELETE) are async at the engine level; results may not be immediately visible after datatalk_execute_sql returns. The chat-path DELETE confirmation flow still applies — see [[sql-execution]].
• File and network access functions (file, s3, url, remote, hdfs, odbc, jdbc, mysql, postgresql) are not supported.
• Cluster and system operations (SYSTEM, KILL QUERY, OPTIMIZE, ATTACH, DETACH) are not supported.
• SQL splitter: generic (no DELIMITER, no PL/SQL, no GO; format / SETTINGS clauses are statement-internal).
• Risk guard: SHOW, DESCRIBE, EXPLAIN are L1; bounded INSERT and safe CREATE TABLE are L2; DROP, TRUNCATE, ALTER, RENAME, GRANT, REVOKE, CREATE USER/ROLE/DICTIONARY are L3. External table functions (remote(), url(), s3(), file(), etc.) are hard reject.
• Schema context: database selector visible (replaces schema selector as ClickHouse has no schema layer). System databases (system, INFORMATION_SCHEMA, _temporary_and_external_tables) are filtered.
• Diagnostics: structured unsupported. EXPLAIN, lock info, pool status, table space, index hints, terminate session, and optimize table are all unsupported.
• ER Inspector: Day-1 dialect_unsupported (no FK constraints in the OLTP sense) — see [[er-tabs]].
• ER Designer: Day-1 dialect_unsupported (table engine decisions not mappable to DataTalk ER DDL contract).

Apache Doris

• Connection kind: apache_doris (alias doris is normalized to apache_doris). OLAP database with MySQL-compatible network protocol.
• Fields: host (FE hostname or IP), port (default 9030 for FE MySQL protocol), username, password, databaseName. Uses MySQL Connector/J driver.
• Chat-path execution: SELECT / INSERT / UPDATE / CREATE / ALTER…ADD run directly via datatalk_execute_sql; DELETE triggers the in-chat confirmationId flow; destructive DDL (DROP / TRUNCATE / ALTER…DROP / GRANT / REVOKE / DENY / KILL / SHUTDOWN / PURGE / SET GLOBAL / INSERT OVERWRITE) returns redirect_to_editor — AI opens a query_editor tab for manual execution. See [[sql-execution]].
• Doris DML may be async for some operations; results from datatalk_execute_sql may not be immediately visible. The standard DELETE confirmation flow still applies — see [[sql-execution]].
• Bulk load operations (LOAD LABEL, ROUTINE LOAD, STREAM LOAD, EXPORT) and cluster management (ADMIN, ALTER SYSTEM, SHUTDOWN, DECOMMISSION) are not supported through the chat path.
• SQL splitter: reuses MySQL splitter (DELIMITER, backtick identifiers, comment handling).
• Risk guard: SHOW, DESCRIBE, EXPLAIN are L1; INSERT, CREATE TABLE, CREATE INDEX, ANALYZE are L2; DROP, TRUNCATE, ALTER, GRANT, REVOKE, CREATE USER/ROLE, LOAD, ROUTINE LOAD, EXPORT, ADMIN, SHUTDOWN are L3.
• Schema context: database selector visible (Doris databases map to MySQL-style catalogs). Schema selector hidden (Doris has no independent schema layer).
• Diagnostics: Day-1 structured unsupported. EXPLAIN, lock info, pool status, table space, index hints, terminate session, and optimize table are all unsupported.
• ER Inspector: Day-1 dialect_unsupported (foreign-key metadata not verified for Doris) — see [[er-tabs]].
• ER Designer: Day-1 dialect_unsupported (Doris DDL has unique distribution / partition syntax).

OceanBase (MySQL-mode)

• Connection kind: oceanbase (no aliases — oceanbase-ce, ob, obcluster, OceanBase are all rejected at ConnectionKind.normalize).
• Day-1 first-class: MySQL-mode only. Oracle-mode (compatibility_mode='oracle') returns dialect_unsupported.
• Fields: host, port (default 2881), username, password, databaseName, compatibilityMode (required), oceanbaseTenant (required), oceanbaseCluster (optional).
• Driver: com.oceanbase:oceanbase-client; URL jdbc:oceanbase://<host>:<port>/<db>; default port 2881.
• Username form <user>@<tenant> or <user>@<tenant>#<cluster> is composed by ConnectionService; AI must NOT fabricate this; structured fields in MCP.
• Chinese aliases (Ant OceanBase, Woqu OceanBase) are recognized in user natural-language input only; they map to canonical kind oceanbase, NOT to mysql.
• Day-1 unsupported: PROCEDURE / FUNCTION / TENANT / OUTLINE / RESOURCE POOL / ALTER SYSTEM / MAJOR-MINOR FREEZE / BACKUP-RESTORE — all classified L3 or dialect_unsupported.
• Chat-path execution: SELECT / INSERT / UPDATE / CREATE / ALTER…ADD run directly via datatalk_execute_sql; DELETE triggers the in-chat confirmationId flow; destructive DDL (DROP / TRUNCATE / ALTER…DROP / GRANT / REVOKE / DENY / KILL / SHUTDOWN / PURGE / SET GLOBAL / INSERT OVERWRITE) returns redirect_to_editor — AI opens a query_editor tab for manual execution. See [[sql-execution]].
• SQL splitter: reuses MySQL splitter (backtick identifier support, DELIMITER handling).
• Risk guard: SHOW, DESCRIBE, EXPLAIN are L1; INSERT, CREATE TABLE, CREATE INDEX are L2; DROP, TRUNCATE, ALTER, GRANT, REVOKE, CREATE OUTLINE, ALTER OUTLINE, CREATE TENANT, ALTER TENANT, DROP TENANT, CREATE RESOURCE POOL/UNIT, ALTER RESOURCE POOL/UNIT, DROP RESOURCE POOL/UNIT, ALTER SYSTEM, MAJOR FREEZE, MINOR FREEZE, BACKUP, RESTORE are L3.
• Schema context: database selector visible (OceanBase databases). System schemas (oceanbase, information_schema, mysql, SYS, LBACSYS) are filtered from target discovery.
• Diagnostics: Day-1 structured unsupported. All 9 hooks (explain_real / index_hints / lock_info / pool_status / table_space / terminate_session / optimize_table / er_inspector / er_designer) return dialect_unsupported.
• ER Inspector: Day-1 dialect_unsupported — see [[er-tabs]].
• ER Designer: Day-1 dialect_unsupported.

StarRocks

• Connection kind: starrocks. OLAP database with native StarRocks JDBC driver.
• Fields: host (FE hostname or IP), port (default 9030 for FE query port), username, password, databaseName (required, StarRocks database within default_catalog).
• Catalog: Day-1 uses default_catalog hardcoded in the JDBC URL (jdbc:starrocks://host:9030/default_catalog.database).
• Chat-path execution: SELECT / INSERT / UPDATE / CREATE / ALTER…ADD run directly via datatalk_execute_sql; DELETE triggers the in-chat confirmationId flow; destructive DDL (DROP / TRUNCATE / ALTER…DROP / GRANT / REVOKE / DENY / KILL / SHUTDOWN / PURGE / SET GLOBAL / INSERT OVERWRITE) returns redirect_to_editor — AI opens a query_editor tab for manual execution. See [[sql-execution]].
• Catalog operations (CREATE/DROP CATALOG), load operations (LOAD LABEL, ROUTINE LOAD, STREAM LOAD, BROKER LOAD), export, cluster management (ADMIN, ALTER SYSTEM), and global variable changes (SET GLOBAL) are not supported through the chat path.
• SQL splitter: reuses MySQL splitter (backtick identifier support, DELIMITER handling).
• Risk guard: SHOW, DESCRIBE, EXPLAIN are L1; INSERT, CREATE TABLE, CREATE INDEX, ANALYZE are L2; DROP, TRUNCATE, ALTER, GRANT, REVOKE, CREATE USER/ROLE/CATALOG, DROP CATALOG, LOAD, ROUTINE LOAD, STREAM LOAD, BROKER LOAD, CANCEL LOAD, EXPORT, ADMIN, SET GLOBAL, SET PASSWORD, KILL, SUBMIT TASK, CANCEL TASK, RENAME, INSERT OVERWRITE are L3.
• Schema context: database selector visible (StarRocks databases within default_catalog). Schema selector hidden.
• Diagnostics: Day-1 structured unsupported. EXPLAIN, lock info, pool status, table space, index hints, terminate session, and optimize table are all unsupported.
• ER Inspector: Day-1 dialect_unsupported (StarRocks DDL has unique distribution / partition syntax) — see [[er-tabs]].
• ER Designer: Day-1 dialect_unsupported (StarRocks DDL has unique distribution / partition syntax).

Trino

• Connection kind: trino. Federated SQL query engine with official Trino JDBC driver.
• Fields: host (Trino coordinator hostname or IP), port (default 8080), username, password (optional), databaseName (Trino catalog, optional).
• Catalog / schema: two-level context model — databaseName maps to Trino catalog, schema maps to Trino schema. Both catalog and schema selectors are visible in the Query Editor. System catalogs (system, memory, jmx) are filtered from target discovery.
• Chat-path execution: SELECT / INSERT / UPDATE / CREATE / ALTER…ADD run directly via datatalk_execute_sql; DELETE triggers the in-chat confirmationId flow; destructive DDL (DROP / TRUNCATE / ALTER…DROP / GRANT / REVOKE / DENY / KILL / SHUTDOWN / PURGE / SET GLOBAL / INSERT OVERWRITE) returns redirect_to_editor — AI opens a query_editor tab for manual execution. See [[sql-execution]].
• Write support is connector-dependent — INSERT, CREATE TABLE AS, UPDATE, DELETE, and DDL may or may not be available depending on the underlying federation connector. Successful execution through datatalk_execute_sql is not a guarantee that other connectors of the same engine accept the same statement.
• Connector caveat: different Trino connectors have different capabilities. Do not claim generic write, ER, or diagnostics support without connector-specific verification.
• SQL splitter: generic (no DELIMITER, no PL/SQL, no GO).
• Risk guard: SHOW, DESCRIBE, EXPLAIN are L1; INSERT, CREATE TABLE, CREATE VIEW, CREATE MATERIALIZED VIEW, UPDATE, DELETE are L2; DROP, TRUNCATE, ALTER, GRANT, REVOKE, CREATE USER/ROLE/CATALOG, CALL, SET SESSION, RESET SESSION, SET PATH are L3.
• Schema context: database selector visible (Trino catalogs via SHOW CATALOGS), schema selector visible (Trino schemas). Catalog + schema are applied via setCatalog() + setSchema().
• Diagnostics: Day-1 structured unsupported. EXPLAIN, lock info, pool status, table space, index hints, terminate session, and optimize table are all unsupported.
• ER Inspector: Day-1 dialect_unsupported — see [[er-tabs]].
• ER Designer: Day-1 dialect_unsupported.
• datatalk_resolve_use_target is required before changing catalog or schema context — see [[connection-management]].
• Trino is NOT a Presto alias. Do not treat trino connections as Presto or route Trino SQL to Presto tools.

Presto

• Connection kind: presto. Federated SQL query engine separate from Trino; uses Presto JDBC driver.
• Fields: host (Presto coordinator hostname or IP), port (default 8080), username, password, databaseName (Presto catalog, optional), and schema (via session data context).
• Catalog / schema: two-level context model — databaseName maps to Presto catalog, schema maps to Presto schema. Both catalog and schema selectors are visible in the Query Editor.
• Chat-path execution: SELECT / INSERT / UPDATE / CREATE / ALTER…ADD run directly via datatalk_execute_sql; DELETE triggers the in-chat confirmationId flow; destructive DDL (DROP / TRUNCATE / ALTER…DROP / GRANT / REVOKE / DENY / KILL / SHUTDOWN / PURGE / SET GLOBAL / INSERT OVERWRITE) returns redirect_to_editor — AI opens a query_editor tab for manual execution. See [[sql-execution]].
• Write support is connector-dependent — INSERT, CREATE TABLE AS, UPDATE, DELETE, and DDL may or may not be available depending on the underlying federation connector. Successful execution through datatalk_execute_sql is not a guarantee that other connectors of the same engine accept the same statement.
• Connector caveat: different Presto connectors have different capabilities. Do not claim generic write, ER, or diagnostics support without connector-specific verification.
• SQL splitter: generic (no DELIMITER, no PL/SQL, no GO).
• Risk guard: SHOW, DESCRIBE, EXPLAIN are L1; INSERT, CREATE TABLE, CREATE VIEW, CREATE MATERIALIZED VIEW, UPDATE, DELETE are L2; DROP, TRUNCATE, ALTER, GRANT, REVOKE, CREATE USER/ROLE, CALL, SET SESSION, RESET SESSION, SET PATH, PREPARE, EXECUTE, DEALLOCATE PREPARE, START TRANSACTION, COMMIT, ROLLBACK are L3.
• Schema context: database selector visible (Presto catalogs), schema selector visible (Presto schemas). System catalogs (system, jmx) are filtered from target discovery.
• Diagnostics: Day-1 structured unsupported. EXPLAIN, lock info, pool status, table space, index hints, terminate session, and optimize table are all unsupported.
• ER Inspector: Day-1 dialect_unsupported — see [[er-tabs]].
• ER Designer: Day-1 dialect_unsupported.
• Presto is NOT a Trino alias. Do not treat presto connections as Trino or route Presto SQL to Trino tools.

Dameng

• Canonical kind: dameng (lower-case). No aliases: dm, dm8, DM, DM8, DM7, dameng7, dameng8, Wuhan Dameng, Dameng are all rejected at ConnectionKind.normalize.
• Driver: com.dameng:DmJdbcDriverX:8.1.x (Maven Central direct; commercial license; offline jar in repo is forbidden by Wave C umbrella §10).
• Driver class: dm.jdbc.driver.DmDriver.
• URL: jdbc:dm://<host>:<port> (server-level; no /<database> suffix; default port 5236).
• databaseName field: reused as the initial schema name (Oracle-precedent; not a separate database). Schema is injected post-connect via SET SCHEMA <name>.
• Day-1 unsupported (returns structured dialect_unsupported):
  • PL/SQL blocks (DECLARE ... BEGIN ... END;)
  • PROCEDURE / FUNCTION / TRIGGER / PACKAGE DDL
  • EXP / IMP CLI utility commands
• Day-1 L3 admin commands (risk label dameng_admin_command): TABLESPACE / USER / ROLE DDL, GRANT / REVOKE, DROP TABLE / VIEW / INDEX / SEQUENCE / SYNONYM. These execute via datatalk_execute_sql like other SQL; explicitly preview the SQL to the user before running because of their admin-level impact — see [[sql-execution]].
• All 7 diagnostics hooks (lock_info, pool_status, table_space, terminate_session, optimize_table, explain_real, index_hints) return structured dialect_unsupported Day-1.
• ER hooks (er_inspector, er_designer) return structured dialect_unsupported Day-1 — see [[er-tabs]].
• Dameng is not an alias of Oracle. AI must not persist or display dameng connections as oracle in any code path.

Apache Hive

• Connection kind: hive. Apache HiveServer2 data warehouse with Hive JDBC driver.
• Fields: host (HiveServer2 hostname or IP), port (default 10000 for binary transport), username, password, databaseName (Hive database).
• Transport / auth: Day-1 supports binary transport only (port 10000) with username/password or username-only authentication. HTTP transport, SSL, Kerberos, ZooKeeper service discovery, custom headers / cookies, and Knox-style deployments are NOT supported.
• Catalog / schema: database-only context — databaseName maps to Hive database. Schema selector is hidden in the Query Editor.
• datatalk_resolve_use_target is required before changing database context via USE — see [[connection-management]].
• Chat-path execution: SELECT / INSERT / UPDATE / CREATE / ALTER…ADD run directly via datatalk_execute_sql; DELETE triggers the in-chat confirmationId flow; destructive DDL (DROP / TRUNCATE / ALTER…DROP / GRANT / REVOKE / DENY / KILL / SHUTDOWN / PURGE / SET GLOBAL / INSERT OVERWRITE) returns redirect_to_editor — AI opens a query_editor tab for manual execution. See [[sql-execution]].
• SQL splitter: generic (no DELIMITER, no PL/SQL, no GO).
• Risk guard: SHOW, DESCRIBE, EXPLAIN are L1; INSERT, CREATE TABLE, CREATE VIEW, ANALYZE are L2; DROP, TRUNCATE, ALTER, GRANT, REVOKE, CREATE USER/ROLE/FUNCTION, LOAD DATA, ADD JAR, TRANSFORM, MSCK, SET, IMPORT, EXPORT are L3.
• Diagnostics: Day-1 structured unsupported. EXPLAIN, lock info, pool status, table space, index hints, terminate session, and optimize table are all unsupported.
• ER Inspector: Day-1 dialect_unsupported — see [[er-tabs]].
• ER Designer: Day-1 dialect_unsupported.

GaussDB

• Canonical kind: gaussdb
• Protocol: PostgreSQL-compatible (centralized mode), password authentication
• JDBC URL: jdbc:postgresql://host:8000/database
• Default port: 8000
• Driver: com.huaweicloud:gaussdbjdbc:v2.0-8.218.0
• SQL splitter: PG splitter (dollar-quoted, PL/pgSQL, stored procedures)
• Risk rules: PG baseline + 5 GaussDB-specific L3 (CREATE RESOURCE POOL, ALTER COORDINATOR, DROP NODE, SHUTDOWN, ALTER SYSTEM SET). All SQL executes via datatalk_execute_sql; explicitly preview the SQL to the user before running these L3 statements because of their cluster impact — see [[sql-execution]].
• Diagnostics: All unsupported (Day-1)
• ER: Unsupported (Day-1) — see [[er-tabs]]
• Day-2: EXPLAIN diagnostics, ER DDL, SSL / TLS
• Day-3: Distributed / DWS mode, Kerberos
• Chat-path execution: SELECT / INSERT / UPDATE / CREATE / ALTER…ADD run directly via datatalk_execute_sql; DELETE triggers the in-chat confirmationId flow; destructive DDL (DROP / TRUNCATE / ALTER…DROP / GRANT / REVOKE / DENY / KILL / SHUTDOWN / PURGE / SET GLOBAL / INSERT OVERWRITE) returns redirect_to_editor — AI opens a query_editor tab for manual execution. See [[sql-execution]].