Skip to workspace
Developer tools / Data
Tools
FREE · PRIVATE BY DEFAULT · NO BACKEND

Database Schema Extractor

Turn SQL definitions and messy structured records into a readable database model. Discover entities, trace relationships, review every guess and export a schema you control.

◎ Visual ER explorer⌘ SQL · JSON · CSV✓ Review inferred links◈ Optional local AI
SCHEMA WORKSPACE

From raw data to clear structure

Ready to analyze
01Source dataLocal files or pasted text
0 characters · Max 4 MiB per file
Extraction and editing stay in your browser. Optional AI downloads public model files only after consent and uses table and column names locally (no row values).
▦0Entities
≡0Columns
⇢0Confirmed links
✧0Suggestions
03

Relationship map

Drag entity cards · Select a field · Zoom and pan

100%
◇Your schema, visualized

Open a file, paste a definition or try a sample.

Solid lines = declared / approved · Dashed = unconfirmed
05

Relationship review

Guesses are not exported until approved

No pending suggestions
No relationships to review yet.
Optional local AI suggestions

Download a free WebGPU model only when you click. Suggestions require your approval.

06

Export & handoff

Deliver a reviewed, portable schema

Export review

Extract a schema to see the output checks.

    WORKFLOW

    Explore complex data without uploading it

    01

    Paste or open

    Use SQL DDL, CSV/TSV, JSON or JSON Lines. Multiple files can be analyzed together within safety limits.

    02

    Verify the structure

    Inspect column types, primary keys and foreign keys. Naming-based links stay suggestions until approved.

    03

    Export your schema

    Download SQL, DBML, Mermaid, editable project JSON or SVG without creating an account.

    Capabilities and limits

    The built-in extractor is a deterministic parser and inference engine, not a neural model. It handles common CREATE TABLE and ALTER TABLE patterns, but it is not a complete SQL dialect parser and does not execute SQL. Type inference from samples is approximate; missing rows, composite keys and complex indexes can require manual review. The optional free on-device AI uses WebGPU, requires a first-use model download and may produce incorrect suggestions. PixelOkay does not receive source data, but loading the optional model contacts external model hosts. No API key or backend is used. CSV rows with missing/extra fields and JSON field-name collisions are rejected rather than silently changed. The export review shows omitted and pending constraints; SQL DDL is illustrative and does not guarantee complete vendor-specific round-tripping.

    LOCAL AI SETUP

    Set up AI on this device

    First, a model needs to be loaded on your device

    Schema extraction and editing work offline. Optional AI suggestions require a free model downloaded from public hosts and run on this device. Only table and column names are passed to the local model.

    1. 1. Check device
    2. 2. Download / cache
    3. 3. Initialize model
    4. 4. Review schema
    Optional on-device relationship suggestions · WebGPU. Download size and model memory depend on the runtime; check device compatibility before loading.

    Device & model resources

    Measurements may be approximate
    Device RAM / model memoryChecking device information…

    Live system RAM, model RAM and GPU memory are not reliably exposed in browsers.

    Browser website storageChecking storage…

    Site-wide estimate; includes other PixelOkay tools and is not RAM.

    ONE ACTION · ALL RECOGNIZED AI MODEL DOWNLOADS

    Free AI model storage across PixelOkay

    Read-only scan of recognized WebLLM and Transformers.js model caches from this entire website origin, including other PixelOkay tools. Unknown caches and browser databases are left untouched.

    Run a scan to see how many model caches can be cleared together.

    Browser-native models and unknown IndexedDB/OPFS or private runtime caches may remain. Unknown application/user storage will not be deleted. Other running tools might redownload caches; close their tabs first.

    Having trouble downloading or loading AI?

    Try refreshing device/storage, Retry Setup, or a different browser. Cache cleanup is optional and must be confirmed; it may require other AI tools to download models again. Never clear all website data without exporting your work.

    Try refreshing device/storage, Retry Setup, or a different browser. Cache cleanup is optional and must be confirmed; it may require other AI tools to download models again. Never clear all website data without exporting your work.

    Offline extraction works without AI. This optional AI action requires a compatible WebGPU browser.
    Recovery and technical details

    If setup fails, review the stage and retry. A WebGPU device does not guarantee sufficient GPU memory.

    Clearing all website data can permanently erase saved projects or drafts across PixelOkay. This tool never clears general site data.

    CROSS-TOOL MODEL CACHE CLEANUP

    Delete recognized AI model downloads?

    Only verified model cache entries/buckets in this browser origin will be removed. Other PixelOkay tools may need to redownload those models. Unknown caches, IndexedDB, OPFS, projects and chats are not targeted. Close other AI tabs first.

    Extract and review a database schema

    Import a supported SQL definition or structured sample, inspect the extracted tables and columns, then review candidate keys and relationships before exporting an editable schema diagram or definition.

    1. Paste or open supported SQL, CSV or JSON input and review parsed tables, column types and any parser warnings.
    2. Inspect candidate primary keys and links. Confirm each suggested relationship against the intended database rules before approving it.
    3. Arrange the entity diagram and export DBML, Mermaid, SQL, SVG or project JSON for further review.

    Example: Load a small customers table and orders sample. Review a proposed link from orders.customer_id to customers.id; verify that customers.id is actually unique before approving it.

    Limits and privacy: The SQL reader supports a subset of DDL and is not a migration runner; unsupported statements may not round-trip. Uniqueness in sample rows does not prove a database constraint. Projects omit source rows and SQL. Optional WebLLM analysis requires WebGPU and downloads runtime/model files only after confirmation; the model prompt uses bounded table and column names, not source row values.

    Related tools