KamehaDB

KamehaDB is a local-first, cross-platform desktop database workspace that connects to PostgreSQL, MySQL, MariaDB, SQLite, SQL Server, Oracle, ClickHouse, DuckDB, MongoDB, Redis, Qdrant, and TigerBeetle — browse schemas, run queries, chat with AI, capture schema history, and generate migrations, all in one Tauri app with a local Node sidecar.

Features

• 12+ engines, one app — SQL, document, cache, vector, and ledger systems side by side: PostgreSQL, MySQL, MariaDB, SQLite, SQL Server, Oracle, ClickHouse, DuckDB, MongoDB, Redis, Qdrant, TigerBeetle
• Local-first — All metadata stored locally in a SQLite database; nothing leaves your machine. No hosted proxy, no telemetry
• AI with real schema context — multi-provider chat (OpenAI, Ollama local + cloud, 9Router) reads your actual DDL, indexes, and constraints to generate SQL, MongoDB pipelines, Redis commands, and Qdrant searches. Streaming responses with stop/cancel
• Read-only MCP server — let Codex, Claude Code, Devin CLI, and OpenCode inspect schemas and run read-only SQL/Mongo queries against opt-in connections over a local 127.0.0.1 endpoint secured with a bearer token
• Schema timeline + diff + migration assistant — capture on-demand snapshots, compare schemas side by side with per-table change cards (columns, indexes), and auto-generate DDL to migrate from one state to the next
• Global search (Ctrl+K) — fuzzy search across connections, tables, columns, open tabs, and quick actions across every connected engine
• Engine-native tooling — PostgreSQL stats and embedded PSQL terminal, MongoDB explorer with chart view, Redis key browser, Qdrant 3D vector map, TigerBeetle accounts/transfers
• Query history with favorites — normalized patterns, duration per group, per-connection
• Charts — bar, line, area, pie views on any query result or Mongo collection (powered by recharts)
• Health monitoring — live status badges with latency tracking; slow/reconnecting/offline states with tooltips
• Schema browser & ER diagram — auto-generated ER graphs via ReactFlow + dagre
• Pin & workspace memory — pinned connections and open tabs persisted to localStorage and restored on refresh
• Schema-aware SQL editor — Monaco with FK-aware JOIN/ON suggestions and autocomplete across tables, columns, functions, keywords
• PostgreSQL pgvector workflows — detect vector columns and indexes, run guided vector/raw search, inspect nearest neighbors, and open a sampled PCA vector map
• Embedded mongosh terminal — interactive mongosh session with full PTY, ANSI colors, resize, and persistence across tab navigation
• Connection URL — paste postgresql://, mysql://, redis://, etc. to auto-fill connection fields

Quick Start

# Install dependencies
pnpm install

# Start dev services
docker compose up -d

# Run the app (sidecar + desktop)
pnpm dev

Then open the desktop app, create a new connection pointing at localhost, and start exploring.

AI Setup

Configure AI providers in the app settings (API Settings page):

Provider	Default Model	Notes
Ollama (local)	llama3.2	Uses http://localhost:11434 by default
Ollama (cloud)	—	Requires base URL and API key
OpenAI	gpt-4o	Requires API key
9Router	Any model	Requires base URL and API key

Read-only MCP setup

KamehaDB runs a read-only Model Context Protocol server on http://127.0.0.1:13979/mcp while the app is open. Local AI clients can inspect schemas and run bounded queries through a dedicated database account that KamehaDB provisions for the selected database.

1. Create a normal connection profile with credentials that can create users and grant read access on its selected database.
2. Open API Settings → MCP Server and choose Create read-only account. KamehaDB grants only read access on that database and stores the generated credential encrypted in its local SQLite metadata store, using an app-managed key file. This works across macOS, Windows, and Linux. Existing accounts created by older Keychain-backed versions must be revoked and recreated. SQLite keeps its existing read-only file connection and does not need a database account.
3. Enable the profile, then copy the generated snippet for Codex, Claude Code, Devin CLI, or OpenCode into its config. Snippets include the endpoint and bearer token.

The MCP listener binds only to loopback, uses a persistent token you can rotate, and stops when KamehaDB closes. MCP queries always use the generated read-only credential; they never fall back to the profile credential. Disabling MCP keeps the database account. Revoke it explicitly before changing the profile's database target or credentials, or deleting the profile. MongoDB account setup requires server authentication to be enabled. If the fixed port is in use, MCP is unavailable until you free it or choose another port — KamehaDB keeps working either way. Aggregations with $out/$merge, write statements, and non-enabled profiles are rejected. See docker-init/MCP-READONLY.md for local database-grant test fixtures.

Connection defaults (Docker)

Engine	Port	User	Password	Database
PostgreSQL	5432	kameha	kameha	kamehadb
MySQL	3306	kameha	kameha	kamehadb
MariaDB	3307	kameha	kameha	kamehadb
Redis	6379	—	—	—
SQL Server	1433	sa	Kameha1!	kamehadb
Oracle	1521	SYS	oracle	ORCLPDB1
ClickHouse	8123	default	default	kamehadb
DuckDB	5432	—	—	—
TigerBeetle	3001	—	—	—
Qdrant	6333	—	—	—

Project Structure

kamehadb/
├── apps/
│   ├── desktop/          # Tauri v2 + React 19 frontend (Vite, Tailwind v4)
│   └── sidecar/          # Hono HTTP server + database adapters + metadata SQLite
├── packages/
│   ├── shared/           # Shared Zod schemas, app state types, adapter contracts
│   └── ui/               # Shared UI primitives
├── landing/              # Next.js marketing & docs site (managed with npm, separate lockfile)
├── docker-compose.yml    # Dev databases
└── docker-init/          # Seed SQL scripts for dev databases

Running Tests

pnpm test

Contributing

See CONTRIBUTING.md for development setup, project conventions, verification requirements, and pull request guidance.

macOS Gatekeeper

Unsigned macOS builds can still be used for local testing, but Gatekeeper may show messages such as "KamehaDB" is damaged and can't be opened or "KamehaDB" cannot be opened because Apple cannot check it for malicious software.

If you only need to test the app on your own Mac, try one of these:

1. Right-click KamehaDB.app, choose Open, then confirm.
2. Remove the quarantine attribute after copying the app to Applications:

xattr -dr com.apple.quarantine /Applications/KamehaDB.app

If the app is still being run directly from the mounted DMG:

xattr -dr com.apple.quarantine "/Volumes/KamehaDB/KamehaDB.app"

You can also disable Gatekeeper globally with sudo spctl --master-disable, but that is broader than necessary and should not be left enabled long-term.

Build for Local Machine

If you see "Target does not exist" error, add the required Rust targets first:

rustup target add aarch64-apple-darwin x86_64-apple-darwin

Then build for your local machine:

# Apple Silicon (M1/M2/M3)
cd apps/desktop && pnpm tauri build --target aarch64-apple-darwin

# Intel Mac
cd apps/desktop && pnpm tauri build --target x86_64-apple-darwin

Linux

# Build for Linux
cd apps/desktop && pnpm tauri build --target x86_64-unknown-linux-gnu

Windows

# Build for Windows
cd apps/desktop && pnpm tauri build --target x86_64-pc-windows-msvc

Tech Stack

• Desktop: Tauri v2, React 19, Vite, Tailwind CSS v4, Base UI (@base-ui/react), TanStack (Query, Store, Table)
• Editor: Monaco (via @monaco-editor/react), custom SQL autocomplete with FK-aware JOIN/ON hints, markdown rendering with react-markdown
• Sidecar: Hono, PostgreSQL (pg), MySQL (mysql2), SQLite (better-sqlite3), MongoDB (mongodb), Redis (ioredis), SQL Server (mssql), Oracle (oracledb), ClickHouse, DuckDB (duckdb), TigerBeetle (tigerbeetle-node)
• Graph: ReactFlow (@xyflow/react), dagre auto-layout
• AI: Multi-provider abstraction (OpenAI, Ollama local/cloud, 9Router)

Roadmap

Currently Supported

• PostgreSQL
• MySQL
• SQLite
• MongoDB
• Redis
• Data export (CSV, JSON, SQL)
• Qdrant (vectordb)
• SQL Server
• Oracle
• ClickHouse
• DuckDB
• Tigerbeetle
• MariaDB

Features

• Query history with favorites
• Data visualization / charts
• Connection health monitoring
• Global search (Ctrl+K)
• Pin / favorites connections
• Workspace tabs memory
• Production safety mode — prevent accidental writes to production
• PostgreSQL backup & restore
• PostgreSQL embedded PSQL tool

In Progress

• Schema change timeline — capture on-demand snapshots, see what changed, and auto-generate DDL to migrate between states
• Migration assistant

Future Ideas

• Collaboration features

License

Apache-2.0 — see LICENSE.