Mako

The AI-native SQL Client.

The Cursor for Data. Connect to any database, query with AI, and build live dashboards -- all from your browser.

Stop wrestling with complex SQL and slow, bloated database tools. Write queries in plain English, get instant results, and turn them into interactive dashboards with cross-filtering and scheduled refresh.

🚀 Why Mako?

A modern SQL client built for the AI era, replacing slow desktop tools with a fast, collaborative, AI-powered experience.

• ✨ AI Query Generation: Write queries in natural language. Our schema-aware AI generates optimized SQL instantly.
  • Replaces: DataGrip, DBeaver, Postico
• 📊 AI Dashboards: Build interactive dashboards from conversation. Cross-filtering, scheduled data refresh, Parquet materialization -- powered by DuckDB in the browser.
  • Replaces: Metabase, Looker, manual BI pipelines
• 🧱 dbt Transforms: Build, run, and schedule dbt Core projects in-app -- file IDE, jobs, run history, lineage, and GitHub sync.
  • Replaces: dbt Cloud
• ⚛️ React Apps: Ask the agent to build live React apps wired to your data through secure, credential-free bindings.
  • Replaces: Lovable, v0, internal-tool builders
• 🕓 Version History: Every console and dashboard save is an immutable snapshot you can browse and restore.
  • Replaces: Lost SQL files, manual backups
• 🖥️ Mako Desktop: Native app that bundles a local agent so localhost databases work out of the box.
  • Replaces: SSH tunnels and bastion hops for local DBs
• 👥 Team Collaboration: Share connections, version-control queries, and work together in real-time.
  • Replaces: Passing credentials around, lost SQL files
• ⚡ Blazing Fast: No Java or Electron bloat. Opens instantly in your browser and runs smooth.
  • Replaces: Slow desktop database tools

📸 Screenshots

AI-powered console — ask in plain English, get a verified query and live results.

Transforms (dbt) — build, run, and schedule dbt Core projects with a file IDE, jobs, run history, and lineage.

Apps — build live React apps wired to your data, rendered in a sandboxed preview.

🔌 Integrations

Databases

Data Connectors

Sync external SaaS data into Mako's data warehouse for querying and dashboards.

🏗️ Architecture

┌─────────────────────────────────────────────────────────┐
│  Frontend (React + Vite)                                │
│  ┌──────────┐ ┌──────────────┐ ┌─────────────────────┐ │
│  │ Console  │ │  Dashboards  │ │   AI Chat (Vercel   │ │
│  │ (Monaco) │ │  (DuckDB +   │ │    AI SDK)          │ │
│  │          │ │   Mosaic)    │ │                     │ │
│  └──────────┘ └──────────────┘ └─────────────────────┘ │
│           ▲          ▲                   ▲              │
│           │     Parquet/Arrow            │              │
│           │     via OPFS cache           │              │
└───────────┼──────────┼──────────────────┼──────────────┘
            │          │                  │
┌───────────┼──────────┼──────────────────┼──────────────┐
│  API (Hono + Node.js)                                  │
│  ┌──────────────────────────────────────────────────┐  │
│  │ Unified Agent (expertise modes: Query /          │  │
│  │   Dashboard / Sync Flow / React App / Transforms │  │
│  │   / Explore, switched via enable_mode)           │  │
│  └──────────────────────────────────────────────────┘  │
│  ┌──────────────┐ ┌───────────────┐ ┌──────────────┐  │
│  │  DB Drivers   │ │  Connectors   │ │  Dashboard   │  │
│  │  (9 drivers)  │ │  (8 sources)  │ │  Engine      │  │
│  │              │ │               │ │  (DuckDB     │  │
│  │              │ │               │ │   + Parquet) │  │
│  └──────────────┘ └───────────────┘ └──────────────┘  │
│                          ▲                             │
│                    ┌─────┴──────┐                      │
│                    │  Inngest   │ (scheduled refresh,   │
│                    │            │  incremental sync)    │
│                    └────────────┘                      │
└────────────────────────────────────────────────────────┘
            │                │
     ┌──────┴──────┐  ┌─────┴──────────┐
     │  MongoDB    │  │  User DBs      │
     │  (metadata, │  │  (PG, BQ, CH,  │
     │   warehouse)│  │   MySQL, etc.) │
     └─────────────┘  └────────────────┘

Key technology choices:

• DuckDB (both server-side via @duckdb/node-api and browser-side via @duckdb/duckdb-wasm): powers dashboard SQL execution, Parquet artifact generation, and in-browser cross-filtering with OPFS caching
• Mosaic (@uwdata/mosaic-core): coordinates cross-filtering across dashboard widgets
• Apache Arrow / Parquet: server materializes query results into Parquet, served to browser as Arrow IPC for zero-copy rendering
• Inngest: event-driven job queues for scheduled dashboard refresh and incremental data sync
• Hono: lightweight, fast HTTP framework for the API
• Monaco Editor: VS Code's editor for the SQL console
• Vercel AI SDK: multi-provider LLM abstraction (OpenAI, Anthropic, Google)

📊 Dashboard Engine

Dashboards are a core feature. The AI agent creates interactive dashboards from natural language:

1. Agent creates a dashboard spec with widgets, layouts, and SQL queries
2. Server materializes query results into Parquet artifacts (stored on filesystem, GCS, or S3)
3. Browser loads Parquet data into DuckDB-WASM, cached in OPFS for instant reloads
4. Mosaic cross-filtering lets users click on one chart to filter all others
5. Inngest cron keeps data fresh with scheduled re-materialization and stale-run detection

Dashboard Artifact Storage

Dashboard materialization stores Parquet artifacts on the backend. Three storage backends:

• filesystem -- default; stores files on local disk
• gcs -- Google Cloud Storage
• s3 -- S3-compatible bucket

DASHBOARD_ARTIFACT_STORE=filesystem

# Optional shared settings
DASHBOARD_ARTIFACT_PREFIX=dashboards
DASHBOARD_ARTIFACT_DIR=/absolute/path/to/artifacts  # filesystem only

Google Cloud Storage

DASHBOARD_ARTIFACT_STORE=gcs
GCS_DASHBOARD_BUCKET=your-bucket-name
DASHBOARD_ARTIFACT_PREFIX=dashboard-artifacts/prod

See the docs for full GCS/S3 provisioning instructions.

🛠️ Quick Start

1. Clone & Install

   bash

   git clone https://github.com/mako-ai/mako.git
   cd mako
   pnpm install
   

2. Configure Environment Copy .env.example (if available) or create .env:

   env

   # Local development connects to the shared `dev` database, an Atlas DB that is
   # refreshed nightly from production. Grab the `dev` connection string from the
   # team vault. (`staging` backs non-migration PR previews; never point local at
   # `production`.)
   DATABASE_URL=mongodb+srv://<user>:<password>@<cluster>.mongodb.net/dev
   ENCRYPTION_KEY=your_32_character_hex_key_for_encryption
   WEB_API_PORT=8080
   BASE_URL=http://localhost:8080
   CLIENT_URL=http://localhost:5173
   

3. Start Services

   bash

   # Start the local notebook Python kernel so notebook `code` cells run
   # locally. Requires the KERNEL_* vars from .env.example in your .env.
   # (The dev database is hosted MongoDB Atlas, set via DATABASE_URL.)
   pnpm run docker:up
   
   # Start the full stack (API + App + Inngest)
   pnpm run dev
   

4. Analyze

   • Open http://localhost:5173 to access the app.
   • Add a Data Source (e.g., Stripe or Close.com).
   • Use the chat interface to ask questions about your data.

🌐 IP Whitelisting

If your database requires IP whitelisting, add the following static IP to your allowlist:

34.79.190.46

This IP is used by Mako's cloud service for all outbound database connections.

💻 Development Commands

🔎 React Performance Profiling

Use React Scan when working on render churn, streaming chat responsiveness, Monaco console performance, or explorer/result table interactions:

pnpm run app:dev:scan

This enables the React Scan Vite plugin via VITE_REACT_SCAN=true and turns on Mako's render-debug logs via VITE_RENDER_DEBUG=true. React Scan highlights components that re-render in the browser, while render-debug logs summarize why hot components such as Chat, ResourceTree, Console, and ResultsTable changed.

Keep React Scan off during normal development. The plugin and debug hooks are gated behind env flags so regular pnpm run app:dev runs without profiling overlays or extra debug logging.

🤝 Community & Support

• Documentation: docs.mako.ai
• GitHub: mako-ai/mako
• Website: mako.ai

Built with ❤️ by the Mako Team. Open Source and self-hostable.