▸ Native · Rust + GPUI · No Electron

A modern database client,
built for performance.

Postgres, MySQL, SQLite, Redis, MongoDB — native on macOS, Windows, Linux.

Built in Rust on GPUI — the same GPU framework that powers Zed. Sub-500 ms cold start. 60 fps on a million rows. No Electron.

Beta coming soon See features

⌘K opens command palette

zolt — analytics_prod production · read-only

weekly_actives.sql schema diff ⌘⏎ run

weektieractive_usersΔ wow

0 rows · — ms 60.0 fps

Concept preview rendered on the web — the real thing is a Rust + GPUI binary. Public beta in ~12 weeks.

ENGINES

postgres 9 → 16

FREE

mysql 5.7 → 8

PRO

sqlite 3.x

FREE

redis 6 → 7

PRO

mongo 5 → 7

PRO

Built in Rust on GPUI.
Nothing else moves like it.

The same GPU framework that powers Zed. Direct Metal / DirectX / Vulkan render path — the result grid is a GPU surface, not a DOM table.

01 — RUST + GPUI

Zero-overhead Rust. Direct GPU render. No browser engine.

zolt is written end-to-end in Rust on GPUI — the same framework that powers Zed editor. Every pixel goes straight from a Rust struct to Metal/DirectX/Vulkan. No Electron, no WebView, no garbage collector pause. Cold start is sub-500 ms; scrolling 1M rows holds 60 fps.

cold start

412 ms

sub-500 ms

scroll fps

60.0

1M rows

frame time

16.6 ms

transform-only

render budget · scroll 1M rows

zolt · Rust + GPUI 16 ms

native, mac-only 38 ms

electron client 84 ms

lower is better — measured on Apple M2, postgres 16, 1,000,000 rows

02 — AI INTEGRATION PRO

Bring your own model. zolt never proxies a single token.

Plug in Ollama, the claude CLI, an Anthropic / OpenAI key, or any MCP server. Your provider bills you directly. We never see your prompts, your responses, or your keys.

ollama

local · llama3.1

claude CLI

anthropic

openai

api key

MCP server

any tool

weekly active users grouped by plan tier ⌘K

-- preview · BYOK · routed to ollama (local)
select date_trunc('week', s.started_at), u.tier,
       count(distinct s.user_id)
from sessions s join users u on u.id = s.user_id
group by 1, 2 order by 1;

● routed to ollama · local · 0 tokens billed by zolt ⌘⏎ insert

Built for the people who already live in psql.

GPU-rendered grid

Virtualized, 1M rows, transform-only scroll. The grid is a Metal/DirectX/Vulkan surface, not a DOM table.

scroll fps · last 3 s ● 60.0 fps

rows

1,000,000

frame

16.6 ms

mem

184 MB

Native on every desktop OS

Same Rust binary class on macOS, Windows, Linux. Identical shortcuts. No 'we ported it' second-class citizens.

macOS

12+ · arm64 · x86_64

→ Metal

Windows

10+ · x86_64

→ DirectX 12

Linux

GNOME · KDE · Wayland

→ Vulkan

Bring-your-own AI

PRO

Pro unlocks the integration layer. Plug in Ollama, the claude CLI, an Anthropic / OpenAI key, or any MCP server. AI usage is billed by your provider — never by zolt.

top 10 repeat customers in q1 via ollama · llama3.1

-- preview · press ⌘⏎ to insert
select u.email,
       count(*) as orders
from orders o
join users u on u.id = o.user_id
where o.placed_at >= '2026-01-01'
group by 1
order by orders desc limit 10;

SSH tunneling, built-in

Define a bastion once. zolt opens, holds, and tears down the tunnel per session. No PuTTY, no autossh dance.

your laptop

bastion

postgres

SSH (key-auth, ed25519) TLS 1.3

Keychain-only credentials

Passwords live in macOS Keychain, Windows DPAPI, or libsecret. Never on disk in plaintext. Ever.

analytics-prod

postgres · admin@db-1.acme.io

password = ••••••••••••

● keychain · macOS

E2E-encrypted team sharing

TEAM

Connection blobs are sealed with age to each member's Ed25519 key. The sync server holds ciphertext and nothing else.

you

teammate

age-encrypt(blob, ed25519:rae)

→ -----BEGIN AGE ENCRYPTED…

Schema diff & DDL gen

Diff two schemas, generate the migration, paste into your repo. No external diff tool.

+ ALTER TABLE users ADD COLUMN tier text;

CREATE INDEX users_email_lower

~ ALTER COLUMN orders.total TYPE numeric(12,2);

− DROP TABLE legacy_sessions;

4 changes · postgres · ready to copy

MCP server mode

PRO

Run zolt as a Model Context Protocol server. Claude Desktop calls query_database and describe_schema directly. Tool-use trace, not a chat about your data.

→ tool_use: describe_schema(connection: "analytics-prod")

← 14 tables, 86 columns

→ tool_use: query_database(sql: "select count(*)…")

← 1 row · 12 ms · returned to claude

● mcp.local:7841 · 2 tools registered

Three things you couldn't do this morning.

zolt — capture 1/3 · palette

monthly active users by week ⌘K

Generate SQL · MAU by week ollama · llama3.1

Open dashboard

Run last query again

-- generated · review before run
select date_trunc('week', started_at) as wk,
       count(distinct user_id) as maus
from sessions
where started_at >= '2026-01-01'
group by 1
order by wk;

events · 1,000,000 rows scanned · scroll position 423,118 ● 60 fps

#

event

payload

ts

user

423,118

page_view

{"path":"/p/0"}

14:00:0

u_23s6

423,119

click

{"path":"/p/73"}

14:01:7

u_23s7

423,120

sign_up

{"path":"/p/146"}