Databases
Eleven engines, one client
What works on each engine, the server versions the test suites run against, and the views made for each one. Every engine is a native Rust driver behind one shared interface.
At a glance
| Engine | Browse & edit | Query tabs | Stop | Transactions | ER diagram | SSH & TLS | Vector tools |
|---|---|---|---|---|---|---|---|
| MariaDB | Yes | Yes | Yes | Yes | Yes | Yes | 11.7+ |
| MySQL | Yes | Yes | Yes | Yes | Yes | Yes | No |
| SQL Server | Yes | Yes | Yes | Yes | Yes | Yes | No |
| PostgreSQL | Yes | Yes | Yes | Yes | Yes | Yes | pgvector |
| Oracle | Yes | Yes | Yes | Yes | Yes | Yes | 23ai |
| MongoDB | Yes | Yes | No | No | Yes | Yes | No |
| Redis / Valkey | Yes | Yes | Yes | No | No | Yes | 8+ |
| SQLite | Yes | Yes | Yes | Yes | Yes | No | No |
| ClickHouse | Yes | Yes | Yes | No | No | Yes | 25.8+ |
| Elasticsearch | Yes | Yes | Yes | No | No | Yes | 8.0+ |
| OpenSearch | Yes | Yes | Yes | No | No | Yes | k-NN |
✓ available, — not available. Redis is edited in its key browser, and ClickHouse edits run as mutations. Vector tools need MariaDB 11.7+, PostgreSQL with pgvector, Oracle 23ai, Elasticsearch 8.0+, OpenSearch with the k-NN plugin, Redis 8 (search module) or ClickHouse 25.8+ (Array columns with a vector_similarity index).
Engine by engine
“Tested with” is the server version the CI contract suites run against. Each card links to the engine's notes in the repository.
-
MariaDB
Tested with 11.8
- What works
- Typed grid with inline edits, query tabs with Stop (
KILL QUERY), transactions across runs, statement timeouts, design view and ER diagram, SSH and TLS. - Vector tools
- On 11.7 and newer: embedding search, column and index health checks, and index recall against exact search.
- Notes
- A Read-only workspace also opens read-only sessions.
-
MySQL
Tested with 8.4
- What works
- Typed grid with inline edits, query tabs with Stop (
KILL QUERY), transactions across runs, statement timeouts, design view and ER diagram, SSH and TLS. - Notes
- Same driver as MariaDB (
mysql_async). The vector tools are for MariaDB 11.7+ only.
-
SQL Server
Tested with 2022
- What works
schema.tablenames (dbois hidden),OFFSET/FETCHpaging, batches that return every result set and row count, query tabs with Stop, transactions, design view and ER diagram, SSH.- Notes
- TLS is on by default. An untrusted certificate shows an “Accept invalid certificates” hint.
-
PostgreSQL
Tested with 13, 17
- What works
- Schemas, materialized views, functions, sequences, types and extensions; arrays,
jsonbanduuidvalues; a Stop that cancels on the server;EXPLAIN; transactions across runs; design view and an ER diagram over all user schemas; SSH and TLS. - Vector tools
- With pgvector:
vectorandhalfveccolumns, HNSW and IVFFlat indexes, search by cosine, L2 or inner product, health checks and index recall. - Notes
- Client key files must be unencrypted.
LISTEN/NOTIFY,COPYstreams and logical replication are not supported.
-
Oracle
Tested with 23ai Free
- What works
- Schemas with tables, views, sequences, synonyms, packages, procedures and functions; edits by key or ROWID; Stop with
OCIBreak; transactions from the toolbar; an ER diagram per schema; SSH and TLS. - Engine views
- PL/SQL source with compile errors, and ORA codes with the error position.
- Vector tools
- On 23ai:
VECTORcolumns, IVF and HNSW indexes,VECTOR_DISTANCEsearch, health checks and recall. - Notes
- Needs Oracle Instant Client 19c or newer. On the first connection the app offers to download Instant Client Basic Light from Oracle (about 60 MB, pinned checksum) or to use a folder with an installed client. TLS is
tcpswith a wallet.
-
MongoDB
Tested with 7, 8.0
- What works
- Collections as tables with BSON-typed columns and edits, JSON filters, Atlas URIs, query tabs that run database commands as JSON, an ER diagram with inferred links between collections, SSH and TLS. Double-click a collection to open its workspace: a query bar in mongosh syntax with completion and history, List, JSON and Table views with edits reviewed before Save, bulk updates and deletes with a preview, and saved queries.
- Engine views
- The collection workspace's sections: Aggregations (stage cards with a preview each, export to mongosh, Node.js or Python), Schema (a sample of up to 10,000 documents, mixed types highlighted), Indexes (usage, create, hide and drop), Explain (a visual plan with index suggestions) and Validation (rules generated from the schema and checked against existing documents).
- Notes
- No transactions, and no Stop for a running command.
-
Redis / Valkey
Tested with 7.4, 8.2
- What works
- Every logical database; keys listed with
SCAN, with type, TTL and memory; an editor for each data type (string, hash, list, set, sorted set, stream); TTL changes; SSH and TLS (rediss://). - Engine views
- Key browser, a console with completion and Stop, and a server view with
INFOand the slow log. - Vector tools
- With the search module (built into Redis 8): indexes with
VECTORfields (FLAT or HNSW) open as tables, with vector search and health checks. Redis 7.4 without the module has no vector tools. - Notes
- Also Valkey and other RESP servers. One standalone server: Redis Cluster and Sentinel are not supported yet.
-
SQLite
Tested with 3.50
- What works
- Open a file (read-only, create,
:memory:) with attached databases; edits by primary key orrowid; Stop by interrupt;EXPLAIN; transactions; design view and ER diagram. - Notes
- SQLite is bundled with the app. A database is a local file, so there is no SSH or TLS. Encrypted databases (SQLCipher) and extensions are not supported.
-
ClickHouse
Tested with 25.8
- What works
- Databases, tables, views, materialized views and dictionaries; keyset paging; streaming query tabs that stop with
KILL QUERY; a design view with the sorting key and skipping indexes; SSH and TLS. - Engine views
- Edits as asynchronous mutations on MergeTree tables, a Parts and mutations view, and a Server view with running queries and the query log.
- Vector tools
Array(Float32),Array(Float64)orArray(BFloat16)columns with avector_similarityindex: vector search and health checks. Generally available in 25.8; older servers need an experimental setting.- Notes
- Uses the HTTP interface. No transactions, and no ER diagram (ClickHouse has no foreign keys); mutations cannot be rolled back.
-
Elasticsearch
Tested with 8.19
- What works
- Indices, data streams and aliases; a document grid typed by the mapping, with structured and Query DSL filters;
search_afterpaging past 10,000 documents; Mapping as the design view; SSH and TLS. - Engine views
- A Dev Tools console (
METHOD /pathwith JSON or NDJSON bodies), plus ES|QL and SQL tabs. - Vector tools
- On 8.0 and newer:
dense_vectorfields, search, health checks and index recall. - Notes
- Elasticsearch 7.10 or newer. No transactions: a save checks each document's version and stops with a conflict if it changed.
-
OpenSearch
Tested with 2.19
- What works
- The same document grid, filters and paging as Elasticsearch; basic auth or a bearer token; SSH and TLS.
- Engine views
- The Dev Tools console and SQL tabs through the SQL plugin (no ES|QL).
- Vector tools
- With the k-NN plugin:
knn_vectorfields, search, health checks and index recall. - Notes
- OpenSearch 1.0 or newer; point-in-time paging past 10,000 hits needs 2.4.
The same on every engine
-
Workspaces
Connections, commands and tabs per project or environment, whatever the engine.
-
Safety rules
Every statement passes one gateway with rules per workspace; a Production workspace starts Read-only.
-
AI agent
The agent reads the schema, writes and explains queries, and opens the right view on any engine. It can learn a database once, so later chats already know it.
-
Connection strings
Paste a connection string for any engine; passwords stay in Credential Manager.
-
MCP server
AI agents such as Claude Desktop, Claude Code or Cursor can use your connections through MCP. Off by default; every write needs approval and every call is audited. MCP guide