PostgreSQL
SQLThe reference engine — pooled connections, full catalog introspection, visual EXPLAIN, ER diagrams, live health and agent mode.
- TRANSPORT
- pg pool · 5432
- REFERENCE
- docs/providers/postgres.md
Open source · MIT · v0.18.2
You created the database.
The editor is already beside it.
LibreDB Studio is the database IDE that deploys next to your data instead of onto your laptop — one browser tab for PostgreSQL, MySQL, MongoDB, Redis and twenty-three more engines, with SSO and audit built in.
Open live demo
Deploy in one click Shared account — nothing you leave in it stays private.
docker run -p 3000:3000 libredb/libredb-studio or npx @libredb/studio — no Docker needed. Running in about ten seconds.
Why it exists
The database took forty seconds. Reaching it took the afternoon.
You create a Postgres on Railway. It is ready before your coffee. Then you want to look inside it — so you expose a port to the internet, or install a desktop client and dig an SSH tunnel, or give up and drive it from a shell.
Databases moved — into Kubernetes, into PaaS platforms, into a customer's VPC. The tools that read them did not. They are still desktop applications: heavy, licensed per seat, built for one database, one laptop, one person who never changes machines.
✕ database exposed to the internet — closed by the first security review ✕ tunnel dropped — reconnect, re-authenticate, start over ✕ works — on one machine, for one person, until they change laptops → https://studio.internal — every browser, every phone on the team
The tool goes to the data. Not the data to the tool.
Take that seriously and it becomes a specification: the editor has to run in a browser, reach a phone, deploy like infrastructure and stay unrestricted. MIT is not generosity — it is a requirement of the architecture.
✕ dumps, replicas, tunnels — terabytes travel to one machine → the data never leaves the network — a URL travels instead
runs in a browserreaches a phonedeploys like infrastructureMIT licensed
How it works
From forty seconds to the whole team
-
01
Create the database
A one-line provision, a Helm release, a managed instance — ready in forty seconds, inside a network your laptop cannot see.
-
02
Deploy Studio beside it
One container next to the database. Same network, nothing exposed to the outside, no SSH tunnel dug. Docker, Helm, an operator, or a one-click template.
-
03
Open a browser tab
The editor is a URL. It reaches your laptop, your teammate's machine, and the phone you are holding during an incident. Nothing to install, ever again.
-
04
Sign the whole team in
OIDC single sign-on, role-based access control, an audit trail of every executed query, and risk analysis before a destructive statement runs.
Twenty-seven engines
One interface.
Different engines.
Relational, document, key-value, graph, vector, time-series, streaming, analytical, search and federated query — the same tree, editor and grid over each of them, and each engine's real limits declared rather than papered over.
What this engine does not do
Nothing held back — every capability in the product ships here, including agent mode.
Stated because the alternative is a feature list that quietly breaks on the fourth engine. Capability flags come from the provider itself — a control that cannot work is hidden, not offered and then failed.
MySQL
SQLPooled connections with information_schema introspection, EXPLAIN plans and performance_schema-backed monitoring.
- TRANSPORT
- mysql2 pool · 3306
- REFERENCE
- docs/providers/mysql.md
What this engine does not do
Agent mode is unavailable, as on every engine but PostgreSQL, SQLite, DuckDB and SQL Server.
Stated because the alternative is a feature list that quietly breaks on the fourth engine. Capability flags come from the provider itself — a control that cannot work is hidden, not offered and then failed.
Oracle
SQLPooled connections with introspection through the ALL_* data-dictionary views and monitoring through the V$ views, each read guarded because they need privileges an app user may lack.
- TRANSPORT
- oracledb pool · 1521
- REFERENCE
- docs/providers/oracle.md
What this engine does not do
Agent mode is unavailable here. Operate still reads the server's own reporting interface.
Stated because the alternative is a feature list that quietly breaks on the fourth engine. Capability flags come from the provider itself — a control that cannot work is hidden, not offered and then failed.
SQL Server
SQLPooled connections with sys.* catalog introspection, DMV-backed monitoring and agent mode under a least-privilege read-only profile.
- TRANSPORT
- mssql pool · 1433
- REFERENCE
- docs/providers/mssql.md
What this engine does not do
The editor has no Explain action: SQL Server's estimated plan is a session mode that a statement prefix cannot express. Only agent mode reads a plan, to admit a statement.
Stated because the alternative is a feature list that quietly breaks on the fourth engine. Capability flags come from the provider itself — a control that cannot work is hidden, not offered and then failed.
Db2 LUW
SQLIBM Db2 for Linux, UNIX and Windows over DRDA, through a Rust driver that needs no IBM client. Schema introspection, server-side paging, inline edits on tables, and RUNSTATS and REORG one table at a time.
- TRANSPORT
- drda · 50000
- REFERENCE
- docs/providers/db2.md
What this engine does not do
No EXPLAIN, no query cancel and no Studio-side timeout in this version, and most monitoring panels are empty on purpose. Db2 for z/OS and for IBM i are out of scope.
Stated because the alternative is a feature list that quietly breaks on the fourth engine. Capability flags come from the provider itself — a control that cannot work is hidden, not offered and then failed.
SQLite
SQLThe embedded engine — a file, no network. Full editor, ER diagrams, EXPLAIN and agent mode.
- TRANSPORT
- file · embedded
- REFERENCE
- docs/providers/sqlite.md
What this engine does not do
There is no server to monitor: health reads file size and pragma statistics only.
Stated because the alternative is a feature list that quietly breaks on the fourth engine. Capability flags come from the provider itself — a control that cannot work is hidden, not offered and then failed.
libSQL
SQLSQLite's dialect over the network — Turso and self-hosted sqld, spoken natively.
- TRANSPORT
- hrana · 8080 (443 with TLS)
- REFERENCE
- docs/providers/libsql.md
What this engine does not do
Monitoring reads only what SQLite pragmas and dbstat answer over Hrana: no uptime, cache hit ratio, slow queries or sessions, because libSQL publishes none.
Stated because the alternative is a feature list that quietly breaks on the fourth engine. Capability flags come from the provider itself — a control that cannot work is hidden, not offered and then failed.
DuckDB
AnalyticsIn-process analytics with full EXPLAIN support and agent mode. Admins can query Parquet and CSV files directly; every other role is confined to the database file.
- TRANSPORT
- file · embedded
- REFERENCE
- docs/providers/duckdb.md
What this engine does not do
Single-writer by design: no sessions to list, so health shows storage, not connections.
Stated because the alternative is a feature list that quietly breaks on the fourth engine. Capability flags come from the provider itself — a control that cannot work is hidden, not offered and then failed.
ClickHouse
AnalyticsHTTP interface with system.* introspection, EXPLAIN plans that show index usage, and query-log monitoring.
- TRANSPORT
- http · 8123
- REFERENCE
- docs/providers/clickhouse.md
What this engine does not do
No foreign keys exist, so ER diagrams show structure without discovered relations.
Stated because the alternative is a feature list that quietly breaks on the fourth engine. Capability flags come from the provider itself — a control that cannot work is hidden, not offered and then failed.
Apache Druid
AnalyticsSQL over HTTP with datasource and segment introspection for streaming analytics.
- TRANSPORT
- http · 8888
- REFERENCE
- docs/providers/druid.md
What this engine does not do
Append-oriented: row editing is disabled rather than offered and then failed.
Stated because the alternative is a feature list that quietly breaks on the fourth engine. Capability flags come from the provider itself — a control that cannot work is hidden, not offered and then failed.
Databend
AnalyticsSelf-hosted Databend and Databend Cloud warehouses over its own HTTP query API, with no driver. Catalogs, databases, tables, views and dynamic tables in the tree, EXPLAIN on a SELECT, and monitoring from system tables.
- TRANSPORT
- http · 8000
- REFERENCE
- docs/providers/databend.md
What this engine does not do
No inline row or object edit: Databend declares no keys, so a row has no identity to update. Transactions are not exposed, and EXPLAIN ANALYZE is never sent.
Stated because the alternative is a feature list that quietly breaks on the fourth engine. Capability flags come from the provider itself — a control that cannot work is hidden, not offered and then failed.
Trino
FederatedFederated SQL — one query across the catalogs you already have, from object storage to RDBMS.
- TRANSPORT
- http · 8080
- REFERENCE
- docs/providers/trino.md
What this engine does not do
Trino queries catalogs; it does not manage a lakehouse. Writes depend on the underlying connector.
Stated because the alternative is a feature list that quietly breaks on the fourth engine. Capability flags come from the provider itself — a control that cannot work is hidden, not offered and then failed.
MongoDB
DocumentDocument browsing, aggregation pipelines and index introspection over the native driver.
- TRANSPORT
- driver · 27017
- REFERENCE
- docs/providers/mongodb.md
What this engine does not do
No SQL translation layer is faked — queries here are MongoDB queries.
Stated because the alternative is a feature list that quietly breaks on the fourth engine. Capability flags come from the provider itself — a control that cannot work is hidden, not offered and then failed.
Couchbase
DocumentSQL++ queries with bucket, scope and collection introspection over the documented REST surfaces.
- TRANSPORT
- http · 8091
- REFERENCE
- docs/providers/couchbase.md
What this engine does not do
Agent mode is unavailable, as on every engine but PostgreSQL, SQLite, DuckDB and SQL Server.
Stated because the alternative is a feature list that quietly breaks on the fourth engine. Capability flags come from the provider itself — a control that cannot work is hidden, not offered and then failed.
Redis
Key-valueKey browsing by pattern, type-aware value views and INFO-backed live monitoring.
- TRANSPORT
- resp · 6379
- REFERENCE
- docs/providers/redis.md
What this engine does not do
No SQL, and none is pretended — the editor speaks commands here.
Stated because the alternative is a feature list that quietly breaks on the fourth engine. Capability flags come from the provider itself — a control that cannot work is hidden, not offered and then failed.
etcd
Key-valueReads and writes a cluster through a subset of the etcdctl command line, with keys grouped by prefix and an admin's compaction, defragmentation and alarm disarm. Kubernetes-aware.
- TRANSPORT
- grpc · 2379
- REFERENCE
- docs/providers/etcd.md
What this engine does not do
Writes under Kubernetes' own prefixes are refused before anything is sent, and Kubernetes secrets, protobuf and encrypted values are withheld on every surface. No agent execution and no MCP.
Stated because the alternative is a feature list that quietly breaks on the fourth engine. Capability flags come from the provider itself — a control that cannot work is hidden, not offered and then failed.
Oxia
Key-valueKey lookups, lists and range scans over an Oxia namespace, written as oxia client commands, with its shards in the tree and every key in the Keys panel.
- TRANSPORT
- grpc · 6648
- REFERENCE
- docs/providers/oxia.md
What this engine does not do
Read-only whatever the toggle says: Oxia has no authorization and usually holds Pulsar's metadata, so writes, watches and sessions wait for a design of their own.
Stated because the alternative is a feature list that quietly breaks on the fourth engine. Capability flags come from the provider itself — a control that cannot work is hidden, not offered and then failed.
LibreDB
Key-valueThe embedded ordered key-value store — a file, no server and no wire protocol. Mapped onto the interface by convention rather than by emulating SQL.
- TRANSPORT
- file · embedded
- REFERENCE
- docs/providers/libredb.md
What this engine does not do
No sessions and no index objects exist, so those two panels are absent with their reason rather than answered empty. Maintenance operations are refused, not faked.
Stated because the alternative is a feature list that quietly breaks on the fourth engine. Capability flags come from the provider itself — a control that cannot work is hidden, not offered and then failed.
Cassandra
Wide-columnCQL editor with keyspace, table and type introspection for wide-column data.
- TRANSPORT
- cql · 9042
- REFERENCE
- docs/providers/cassandra.md
What this engine does not do
No joins and no EXPLAIN — the grid respects partition-key query rules instead of hiding them.
Stated because the alternative is a feature list that quietly breaks on the fourth engine. Capability flags come from the provider itself — a control that cannot work is hidden, not offered and then failed.
Neo4j
GraphCypher over Bolt, with node labels, relationship types, indexes and constraints in the tree, and a Graph tab that draws the nodes and relationships a result returns.
- TRANSPORT
- bolt · 7687
- REFERENCE
- docs/providers/neo4j.md
What this engine does not do
Read-only by construction: a read policy, the server's own classification and a READ session stand between a statement and a write. No EXPLAIN or PROFILE, no APOC and no LOAD CSV.
Stated because the alternative is a feature list that quietly breaks on the fourth engine. Capability flags come from the provider itself — a control that cannot work is hidden, not offered and then failed.
Milvus
VectorCollections, exact counts, queries and vector searches over the gRPC API, typed as Milvus's own REST requests, with integers above 2^53 kept exact.
- TRANSPORT
- grpc · 19530
- REFERENCE
- docs/providers/milvus.md
What this engine does not do
Reads only — an admin's Load and Release are the one state change, and a collection is never loaded implicitly. Server-side functions, model rankers and calls to an embedding provider are refused.
Stated because the alternative is a feature list that quietly breaks on the fourth engine. Capability flags come from the provider itself — a control that cannot work is hidden, not offered and then failed.
Qdrant
VectorCollections, point reads, scrolls, counts, facets and queries over dense, sparse and multivector data, written as Qdrant's own REST requests.
- TRANSPORT
- http · 6333
- REFERENCE
- docs/providers/qdrant.md
What this engine does not do
Writes nothing — every request is one of seventeen reads — and refuses every server-side inference input except the local BM25 model.
Stated because the alternative is a feature list that quietly breaks on the fourth engine. Capability flags come from the provider itself — a control that cannot work is hidden, not offered and then failed.
Elasticsearch
SearchElasticsearch SQL over indices, with mapping introspection and cluster health.
- TRANSPORT
- http · 9200
- REFERENCE
- docs/providers/elasticsearch.md
What this engine does not do
Search engines are query-and-browse: no row editing, no ER diagrams.
Stated because the alternative is a feature list that quietly breaks on the fourth engine. Capability flags come from the provider itself — a control that cannot work is hidden, not offered and then failed.
OpenSearch
SearchThe same query-and-browse surface as Elasticsearch, against OpenSearch clusters.
- TRANSPORT
- http · 9200
- REFERENCE
- docs/providers/opensearch.md
What this engine does not do
Search engines are query-and-browse: no row editing, no ER diagrams.
Stated because the alternative is a feature list that quietly breaks on the fourth engine. Capability flags come from the provider itself — a control that cannot work is hidden, not offered and then failed.
Prometheus
Time-seriesPromQL over the HTTP API, with the metric, rule and target browser beside it.
- TRANSPORT
- http · 9090
- REFERENCE
- docs/providers/prometheus.md
What this engine does not do
Read-only: Studio calls the read APIs only, so there is nothing to write back.
Stated because the alternative is a feature list that quietly breaks on the fourth engine. Capability flags come from the provider itself — a control that cannot work is hidden, not offered and then failed.
InfluxDB (InfluxQL)
Time-seriesInfluxQL over the v1 query API on InfluxDB 1.x, 2.x and 3, with databases, measurements, tag keys and field keys in the tree.
- TRANSPORT
- http · 8086
- REFERENCE
- docs/providers/influxdb.md
What this engine does not do
Read-only whatever the toggle says, and no Flux: Flux is a server-side language, and on a read-only token it still made the server issue HTTP requests, so no read-only Flux policy can be sound.
Stated because the alternative is a feature list that quietly breaks on the fourth engine. Capability flags come from the provider itself — a control that cannot work is hidden, not offered and then failed.
InfluxDB 3 (SQL)
Time-seriesInfluxDB 3 in SQL over /api/v3/query_sql, the DataFusion dialect its server runs: the tables of one database, their columns and their newest rows.
- TRANSPORT
- http · 8181
- REFERENCE
- docs/providers/influxdb3.md
What this engine does not do
Read-only, with no maintenance. Cloud Serverless and Cloud Dedicated serve SQL only over Arrow Flight, which Studio does not use; reach them with the InfluxQL type.
Stated because the alternative is a feature list that quietly breaks on the fourth engine. Capability flags come from the provider itself — a control that cannot work is hidden, not offered and then failed.
Apache Kafka
StreamingTopic, group and broker browser that reads by offset or by time.
- TRANSPORT
- kafka · 9092
- REFERENCE
- docs/providers/kafka.md
What this engine does not do
Read-only: it reads records, it does not produce them.
Stated because the alternative is a feature list that quietly breaks on the fourth engine. Capability flags come from the provider itself — a control that cannot work is hidden, not offered and then failed.
The claim is the span, not a count — twenty-seven engines, one interface, plus LibreDB, the embedded store the playground runs on. Twenty-seven more speak one of these wire protocols and connect through the same drivers, each measured against a live instance: fifty-four named engines in all. Per-engine detail lives in the provider docs.
The product
A serious IDE, not an admin panel
Monaco under the hood, a virtualized grid over millions of rows, and the panels a DBA actually opens.
studio — live tour query.sql query.sql query.sql studio — on a phone postgres · prod-eu
⌘K filter tables…
▸ public
▤ orders 184k
id uuid PK
user_id FK
status text
total_cents int4
▤ users 12.4k
▤ order_items 512k
▤ products 2.1k
▤ payments 96k
-- checkout latency, last 24 h
SELECT o.id, u.email, o.total_cents, o.created_at
FROM orders o
JOIN users u ON u.id = o.user_id
WHERE o.status = 'pending'
ORDER BY o.created_at DESC
LIMIT 50; Results History Saved 50 rows · 14 ms
idemailtotal_centscreated_at
b91c-4a… ada@acme.dev 12400 2026-08-27 09:14:02
4e2a-90… kim@acme.dev 8950 2026-08-27 09:11:47
77d0-c3… noor@acme.dev 41200 2026-08-27 09:08:19
c3ab-11… leo@acme.dev 3300 2026-08-27 08:59:55
90ff-7e… mia@acme.dev 27860 2026-08-27 08:57:31
▤ users
id uuid PK
email text
▤ orders
id uuid PK
user_id FK
status text
▤ order_items
order_id FK
product_id FK
qty int4
▤ products
id uuid PK
price_cents int4
3 foreign keys · discovered from the schema · export SVG / PNG
Active sessions
24
Cache hit ratio
99.2%
Database size
4.2 GB
Slowest queries · last hour
SELECT … FROM order_items JOIN products ON … 1840 ms
UPDATE orders SET status = … WHERE created_at < … 902 ms
SELECT count(*) FROM payments WHERE settled = false 488 ms
read from the engine's own reporting interface · sessions can be terminated from here
The incident does not wait
The same editor reaches the phone in your pocket
Dedicated mobile navigation and card views for results — because the query that matters is often the one you run on call, away from a laptop.
What ships
Everything a database IDE owes you
- 01
Schema-aware SQL editor
Monaco under the hood — table and column completion, formatting, snippets, ⌘⏎ to run.
- 02
Visual EXPLAIN
Plan trees for PostgreSQL, MySQL, DuckDB, ClickHouse, Trino and more — scan types, joins and costs, drawn.
- 03
ER diagrams
Foreign-key graphs discovered from the schema. Pan, filter, and export as SVG or PNG.
- 04
AI that knows your schema
Explain a query or draft one in plain language — Gemini, OpenAI, Ollama or any compatible endpoint. Generated SQL is yours to read before it runs.
- 05
Live health dashboard
Sessions, slow queries, cache hit ratios and storage — read from each engine's own reporting interface.
- 06
The database agent
State an objective; the agent runs read-only, budgeted SQL and writes a report where every claim cites a result.
Open source
Nothing sits behind an enterprise wall
LibreDB Studio is MIT because it has to go everywhere. Nothing is sold: no paid edition, no hosted offering, and no capability held back from the MIT build to create a reason to pay.
| Capability | LibreDB Studio | Typical community editions |
|---|---|---|
| OIDC single sign-on | in the MIT build | paid editions only |
| AI assistant | in the MIT build | most providers and features in PRO editions |
| MongoDB and Redis | in the MIT build | absent from community driver sets |
License and feature scope only.
- 100% line coverage, enforced as a CI gate
- 27 engines — SQL, NoSQL, analytics, graph and vector
- 46 distribution channels, 39 of them live
- MIT license. All of it, nothing held back
Deploy
It installs like infrastructure, because it is
A container, a Helm chart, a certified Rancher listing, a one-click template, or an npm package embedded inside your own product.
docker run -p 3000:3000 libredb/libredb-studio
# zero-config first run — secrets generated at boot, admin password printed once
npx @libredb/studio
# no Docker needed — binds to 127.0.0.1 by default
helm install libredb oci://ghcr.io/libredb/charts/libredb-studio
# also a certified chart in Rancher Partner Charts
npm i @libredb/studio
# mount the editor inside the product that created the database
Railway →Dokploy →CapRover →DigitalOcean →Sealos →
# one click on the platform that already runs your database
Bring the editor to the data.
Open the live demo, or have your own Studio running next to your database in about a minute.
Open live demo
Deploy Studio Shared account — nothing you leave in it stays private.