// sqlly vs tusk

SQLly vs Tusk

Tusk is a free, MIT-licensed database client that first shipped in September 2026: keyboard-driven, 20 engines, an AI panel that writes queries but never runs them. It’s built on the same stack as SQLly - Rust, drawn on the GPU by GPUI, with the gpui-component widget kit - so this is a comparison of two apps with the same foundations and different priorities. Tusk goes wide on engines and keyboard flow; SQLly goes deep on the SQL editor, guardrails, and DBA tools.

Last reviewed: November 2026

The short version

SQLly

A deep SQL workbench

One Rust codebase on GPUI for macOS, Windows, and Linux. Thirty engines, IntelliSense that reads your data, results that spill to disk, layered safety guardrails, plans, schema compare, and a full set of script-first admin tools. Free for hobbyists; paid licenses for everyone else.

Tusk

A free, fast, keyboard-first client

Also Rust on GPUI. Twenty engines including Snowflake, BigQuery, MongoDB, Cassandra and DynamoDB, a command palette for every action, an editable grid with undo, structure and index editors, 30+ themes, and an AI panel driven by Claude, Codex or Gemini CLI. MIT, no account, no telemetry. Very young: version 0.2.1, first released on 28 September 2026.

Feature by feature

SQLly’s column uses the same honest status badges as the features page and the docs. For Tusk we list what its README, release notes, and source tree state as of version 0.2.1.

Feature comparison of SQLly and Tusk
Capability SQLly Tusk
Runs on macOS, Windows & Linux YesmacOS is signed and notarized; Windows and Linux are a supported beta, on amd64 and arm64. YesmacOS 14+ (signed and notarized), Windows 10/11 on x64 and ARM64 (not code-signed yet), and Linux x86_64 on Wayland as .deb, .rpm and Arch packages.
Intel Macs NoApple Silicon only. NoApple Silicon only.
How the app is built GPUIOne Rust codebase on GPUI and gpui-component, rendered through Metal, Direct3D or Vulkan. How it works GPUIThe same stack: Rust on GPUI and gpui-component (through gpui-kit), with small vendored patches.
Download size InfomacOS dmg 65.6 MB, Windows x64 zip 60.9 MB, Linux x64 tarball 76.2 MB. InfomacOS dmg 85.3 MB (bundles PostgreSQL client tools and two language servers), Windows x64 zip 39.6 MB, Ubuntu .deb 54.5 MB.
Free to use Free tierTwo saved servers, one cloud account, and a reminder to pay about every five hours. Pricing YesFree with no account and, per its README, no telemetry.
Open source NoLifetime licenses carry a source-escrow promise if SQLly is ever discontinued. YesMIT on GitHub.
PostgreSQL, MySQL, MariaDB, SQLite, SQL Server YesPlus Azure SQL (Entra sign-in in progress). Full list Yes
DuckDB, libSQL, Oracle, ClickHouse YesDuckDB is bundled. Per-engine coverage YesOracle goes through the pure-Rust oracle-rs crate.
Redshift, Greenplum, Vertica, Cloudflare D1 No Yes
Snowflake & BigQuery Planned Yes
MongoDB, Cassandra, DynamoDB No Yes
Redis YesTyped key editing, Cluster and Sentinel, ACLs, pub/sub, memory analyzer, slow log; Valkey, KeyDB, DragonflyDB, ElastiCache and MemoryDB too. Redis YesListed as an engine; we didn’t find Redis-specific tools documented.
SSH tunnels & TLS YesSSH hops, SOCKS and HTTP proxies, Kubernetes, and every TLS mode. TLS YesSSH with a password or key, and any SSL mode.
Import connections from other tools YesDataGrip, DBeaver, pgAdmin, Azure Data Studio, DbGate, TablePlus, Navicat, Redis Insight, pg_service.conf. Compatibility PartialTablePlus (with Keychain passwords, on macOS), a Docker Compose file, or a connection URL.
Command palette & themes - YesEvery action and its shortcut in one palette, quick-open for tables, and more than 30 themes including Catppuccin, Tokyo Night, Gruvbox and Nord.
SQL completions YesSQLly’s own parser: alias- and scope-aware, JOIN ON from foreign keys, real column values in WHERE, GROUP BY mistakes flagged. IntelliSense YesFrom bundled language servers (postgres-language-server for Postgres, sqls for most others), with diagnostics; live-catalog completions elsewhere. Not bundled on Windows yet.
Edit table data YesA row editor generated from the table’s own metadata. Row editor YesTyped cell editors, pending edits saved in one transaction, and full undo and redo.
Edit structure & indexes YesDesigners and Edit Structure. Designers YesColumns, types, defaults, comments and indexes in a grid, with each engine’s own DDL and a warning before a change the engine can’t make in place.
Large results YesRows stream and spill to disk past 50,000 rows or 64 MiB; memory stays flat at 50 million rows. Streaming PartialA virtualized grid that pages table data 1,000 rows at a time.
Pivot, charts & result diff YesPivot, charts, conditional formatting, computed columns, and diffs on any result. Pivot No
Import & export data YesCSV, JSON, SQL, and Excel with live formulas and a pivot sheet. Export PartialExport CSV, JSON or SQL; import CSV. Copy rows as TSV, CSV, JSON, INSERT, UPDATE or Markdown.
Execution plan viewer In progressPlain-language findings, tree, and diagram across nine engines. Query plans -
ER diagrams YesExport to SVG, Mermaid, or DBML. Relationship Graph -
Schema & data compare YesRisk-graded schema diffs across engines and table diffs, with sync and rollback scripts. Schema Compare -
Session & activity monitoring YesActivity Monitor with blocking chains, expensive queries and waits, plus a Server Dashboard. Activity Monitor PartialA process list with cancel and kill on every engine. PostgreSQL role management too.
Backup & restore YesAs reviewable scripts. Backup / Restore PartialPostgreSQL only, through pg_dump and pg_restore (bundled on macOS).
Guardrails in the editor YesSELECT-only connections, no UPDATE/DELETE without WHERE, auto-rollback, tenant checks, inherited per environment. Safety PartialSafe Mode asks before saving grid edits or running a writing statement; it’s off by default. Status colors such as red for production.
AI assistant YesLocal models (Ollama, MLX), your own endpoint, or opt-in cloud providers; no default provider. On-device AI YesClaude, Codex or Gemini CLI over the Agent Client Protocol, under your own CLI account. The agent writes SQL into a tab and never runs it.
MCP server for AI agents Yessqlly-cli mcp for any agent: schema tools and a read-only SELECT that is always rolled back; writes come back as proposals. Read-only enforcement PartialA built-in MCP server feeds the in-app agent: list and describe tables, read or write the query tab. It doesn’t run queries for the agent.
Automatic updates - PartialSigned self-updates on macOS (and Homebrew); manual downloads on Windows and Linux.

A dash means we couldn’t confirm it from Tusk’s own materials (or, for SQLly’s column, that we don’t make the claim here) - check the repository for the current state.

Where SQLly is different

Same UI framework, so the differences are in what each team chose to build on top of it.

Its own SQL brain

Tusk hands completions to external language servers, which is a sensible way to cover many engines quickly. SQLly parses the SQL itself, so it can pull a column’s real values into a WHERE, write a join’s ON clause from actual foreign keys, and keep a 0.5 MB script at about 5 ms per keystroke.

IntelliSense →

Results that don’t run out of memory

Query results stream and spill to disk past 50,000 rows or 64 MiB, so fifty million rows use about as much memory as fifty. On top of the grid sit pivots, charts, conditional formatting, computed columns, and result diffs.

Spill-to-disk results →

Guardrails that are on by design

Read-only connections that refuse file-reading and shell-escape tricks, a hard stop on UPDATE/DELETE with no WHERE, auto-wrapped transactions, and prod painted red - set once on a client, project or environment and inherited by every connection under it.

Safety & guardrails →

A DBA toolbox

Activity Monitor, Query Insights, Extended Events, SQL Agent, an index analyzer, a test data generator, disk usage, backup and restore, and schema and data compare. Anything destructive opens as SQL in a tab for you to read first.

Tools →

Agents outside the app, kept read-only

Both apps refuse to let an agent change data. SQLly’s MCP server also lets any agent run a SELECT, inside a transaction the database keeps read-only and then rolls back; a write comes back as a proposal you confirm, and production connections are refused.

Read-only enforcement →

Moving data between engines

Data Transfer copies tables from one engine to another with type mapping, and schema compare works across engines too, with sync and rollback scripts.

Data Transfer →

When Tusk is the better choice

Tusk is probably the better fit if:

  • You use Snowflake, BigQuery, Redshift, Vertica, Cloudflare D1, MongoDB, Cassandra or DynamoDB. Tusk connects to all of them today; SQLly doesn’t connect to any (Snowflake and BigQuery are planned).
  • Open source, or free at any company size, is a requirement. Tusk is MIT with no account and no paid tier. SQLly’s source isn’t open, and its free tier isn’t for organizations with more than 100 employees.
  • You live on the keyboard. Tusk’s command palette lists every action with its shortcut, and it ships more than 30 editor-style themes.
  • You already pay for Claude, Codex or Gemini CLI. Tusk drives the agent you already use, under that CLI’s own account, instead of asking you to set up a provider.
  • You want a smaller Windows download. Tusk’s Windows x64 zip is 39.6 MB against SQLly’s 60.9 MB. (On macOS it’s the other way round: 85.3 MB against 65.6 MB, partly because Tusk bundles the PostgreSQL client tools and its language servers.)
  • You mostly browse and edit tables. If you don’t need plans, compare, or admin tools, a free and fast client is the simpler choice.

One honest caveat: Tusk is weeks old. Version 0.2.1 came out on 30 September 2026, the Windows build isn’t code-signed yet, and Linux needs a Wayland desktop. It’s moving fast, so check its repository for the current state.

Questions

Is Tusk free?

Yes. Tusk is free, open source under the MIT license, and needs no account; its README says it sends no telemetry. There is no paid tier. SQLly is free for hobbyists, with a one-time $100 SQLly 1.x license and a $200 Lifetime license for everyone else.

Are Tusk and SQLly built the same way?

Largely, yes. Both are Rust apps drawn on the GPU with GPUI, the UI framework from the Zed editor, and both use the gpui-component widget library with small local patches. They share several drivers too, such as tiberius for SQL Server and the DuckDB crate. The main difference is in the editor: Tusk gets completions from bundled language servers (postgres-language-server and sqls), while SQLly parses SQL and builds its completions itself.

Which databases do SQLly and Tusk support?

Both connect to PostgreSQL, MySQL, MariaDB, SQLite, DuckDB, libSQL, SQL Server, Oracle, ClickHouse, Redis and CockroachDB. Tusk adds Amazon Redshift, Greenplum, Vertica, Cloudflare D1, Snowflake, BigQuery, MongoDB, Cassandra and DynamoDB - 20 engines in all. SQLly adds Azure SQL (Entra sign-in in progress) and documented Redis- and Postgres-compatible services such as Valkey, ElastiCache, Neon and Supabase - 30 engines in all.

// try it yourself

Point both at the same database

Same framework, different priorities - the quickest way to choose is to run your own queries in each. SQLly is free on macOS, Windows, and Linux, with no account required.