NATIVE MACOS · STRUCTURED DATA STUDIO

The Mac-native home for your structured data.

JSON, CSV, Markdown, YAML, TOML, and XML — opened, validated, and transformed in one native app. Tree-sitter parsing, on-device AI, and the macOS document model done right. Everything stays on your Mac.

Download for macOS View on Mac App Store

~/projects/atlas/config — Parselet

config.json

schema.json•

users.csv

README.md

Cargo.toml

{

"$schema": "https://json-schema.org/draft/2020-12/schema",

"title": "Atlas user record",

"type": "object",

"required": ["id", "email"],

"properties": {

"id": { "type": "string", "format": "uuid" },

"email": { "type": "string", "format": "email" },

"plan": { "enum": ["free", "pro", "team"] },

"createdAt": { "type": "string", "format": "date-time" },

"trialDays": { "type": "integer", "minimum": 0, "maximum": 90 },

"verified": { "type": "boolean", "default": false }

}

• Valid · JSON Schema 2020-12 UTF-8LF2-space

Ln 13, Col 64 tree-sitter • 4ms

Built for macOS 15+ Universal binary Notarized Sandboxed

JSON· CSV· Markdown· YAML· TOML· XML· JSON· CSV· Markdown· YAML· TOML· XML·

SIX FORMATS

One editor. Six grammars.

Each format gets a tree-sitter grammar, a deterministic parser, and a supplementary view tuned to its shape — not a single text box pretending to be six.

json

.json · .jsonc

Config, API responses, app state. The default exchange format that won.

{

"name": "parselet", "version": "1.0.4",

"depends": ["tree-sitter", "mlx"]

}

csv

Tabular data without the headache of a spreadsheet rewriting your dates.

Prose with just enough structure. README, docs, notes, anything.

# Release notes

> A native editor for structured data.

- incremental parsing ✓

yaml

.yml · .yaml

Configuration that humans actually read and write.

Modern config — Cargo, pyproject, and friends.

Enterprise interchange, SVG, RSS, configs that refuse to die.

<key>CFBundleID</key>

<string>com.parselet</string>

Stays on your Mac. No accounts. No telemetry. No cloud round-trips. The data you open never leaves the device.

INCREMENTAL PARSING

A 50 MB JSON file shouldn't make your editor stutter.

Parselet uses tree-sitter for syntax highlighting and structural parsing, integrated with Neon for incremental updates. Type a character, and only the affected nodes re-parse — the rest of the tree stays where it is. The editor is built on AppKit and TextKit 1, the same text engine Apple uses for documents that need to scale. Open large config files, generated API responses, and exported data dumps without watching a spinner.

Six tree-sitter grammars, one Neon coordinator, zero web-view detours.

huge_dump.json — 51.4 MB

… 1,247,300 more

{ "id": "u_8af2c", "plan": "pro", "events": 1842 },

{ "id": "u_8af2d", "plan": "free", "events": 7 },

{ "id": "u_8af2e", "plan": "team", "events": 2210 },

{ "id": "u_8af2f", "plan": "pro", "events": 488 },

{ "id": "u_8af30", "plan": "pro", "events": 93 },

{ "id": "u_8af31", "plan": "free", "events": 0 },

… 312,418 more

• Parsed 1,559,718 lines

tree-sitter incremental

Inspector — Inference

Tier 1 · Apple Foundation Models

Instant. No download.

Built into macOS 26 and later. Used for parse-error explanations on small contexts. Runs in the OS-managed model process. Apple's compute, your hardware, no cloud.

macOS 26+ · ~3K characters per request · Greedy decoding

Tier 2 · MLX Swift

Deeper reasoning for larger contexts.

A 4-bit quantized Qwen 3.5 4B model, downloaded on first use (~3.2 GB). Runs locally on your Mac via Apple's MLX framework. Used for summarizing files larger than Tier 1's context window. Loaded on demand, released when idle.

Apple Silicon · ≥16 GB unified memory · 32K guaranteed context

ON-DEVICE AI

Two AI tiers. Both run on your Mac.

Parselet's AI features are designed around a single rule: nothing about your data ever leaves the machine. There is no API key to enter, no account to create, no request that goes anywhere. The only thing that travels over the network is the one-time download of the Tier 2 model, and only if you ask for it.

Don't have the hardware? The editor works fully without AI. The AI is a layer, not the product.

A REAL MAC DOCUMENT

Plays by macOS document rules.

Parselet is built on NSDocument, the same document model that ships with TextEdit, Pages, and every other native macOS document app. That means autosave works, Versions works (yes, including Browse All Versions), file associations work, multi-window and multi-tab work, and QuickLook previews show your files with proper syntax highlighting before you even open them. None of this is custom UI — it's macOS doing its job because Parselet wired up the document model the way Apple intended.

·Autosave on by default
·File · Revert To · Browse All Versions
·File associations for all six formats
·QuickLook previews with syntax highlighting
·Multi-window and multi-tab via standard ⌘N and ⌘T
·Recent files menu

Versions — schema.json

Now — 14:42

"verified": { "type": "boolean",

"default": false }

"trialDays": { "max": 90 }

Yesterday — 09:11

"verified": { "type": "boolean" }

"trialDays": { "max": 30 }

Autosaved · 8 versions retained

MENU BAR INTEGRATION

Real menus. Real shortcuts. No web-app pretending.

Parselet ships with a full native menu bar — not a stripped-down pseudo-menu. Standard macOS conventions you already know are wired correctly: cut, copy, paste, find, undo, redo, hide, show all. On top of that, Parselet adds a small set of operations that earn their menu-bar slots:

Parselet — Format menu

Differentiators

File · New from Format

Six explicit starting points (json, csv, md, yaml, toml, xml). Cmd-N defaults to your last choice.
File · Convert To

Deterministic cross-format conversion via real parsers, never AI. JSON ↔ YAML, JSON ↔ TOML, YAML ↔ TOML, CSV → JSON.
Format · Prettify (⇧⌘F) and Minify

Per-format capability. The menu shows what's supported for the current document.
Format Switcher toolbar dropdown

Reinterpret the current document as a different format without modifying the file. View a .json as YAML; switch back; the file is untouched.

Standard cluster

Standard Edit menu (cut, copy, paste, find, undo, redo)
Standard File menu (new, open, save, save as, revert to)
Standard Window menu (next tab, previous tab, merge all windows)
Standard App menu (preferences, hide, hide others, show all)

Behaves like the Mac apps you already trust, because under the chrome it is one.

PRICING

One price. No subscription. Yours forever.

No subscription. Buy it once. All future updates included.

Parselet

One-time

Yours forever.

$29.99 one-time

Buy it once through the Mac App Store. The app keeps working, and future updates are included for life. There is no subscription, trial, upgrade SKU, or version paywall.

Download

✓Full app, all six formats
✓Both AI tiers (hardware permitting)
✓Use on every supported Mac

Available on the Mac App Store.

QUESTIONS

Things people ask before they buy.

Because the app you bought should keep working. Subscriptions make sense when you're paying for a service that runs on someone else's hardware. Parselet runs on yours, and future updates are included.

Yes. Tier 1 uses Apple's Foundation Models, which ship with macOS 26 and run entirely on-device. Tier 2 downloads a 4-bit quantized model on first use (about 3.2 GB) and runs locally on your Apple Silicon Mac. After the one-time Tier 2 download, no network is involved. Nothing about your data is ever uploaded.

The editor runs on macOS 15 (Sequoia) and later. AI features require Apple Silicon (M1 or newer) with at least 16 GB of unified memory. On Intel Macs and 8 GB Apple Silicon Macs, the editor works fully — only the AI features are unavailable.

It doesn't replace your code editor. It replaces the ad-hoc tools you reach for outside it: web-based JSON formatters, Numbers for opening CSVs, online YAML linters, format converters, parser-error explainers. If you write code in VS Code or Xcode, keep doing that. Open the data files those apps consume in Parselet.

Tree-sitter incremental parsing keeps editing responsive across file size. 50 MB JSON files and 100k-row CSVs are normal territory. Beyond that, the bottleneck becomes disk I/O and RAM, not the parser.