Notepatra is a native Qt5 + QScintilla code editor with an original C++17/Rust core, built for the AI era. It runs on Linux, macOS Apple Silicon, and Windows, has a ~12 MB bare native executable (~12.6 MB on Linux x64), supports 238 file types with 82 language lexers, and includes AI-powered formatters that fix broken JSON, HTML, and brackets. The AI panel ships six dropdown backends — two local (Ollama, llama.cpp) and four cloud (Ollama Cloud, OpenRouter, OpenAI, Azure OpenAI) — and the llama.cpp entry also accepts a user-configured base URL for any OpenAI-compatible server.

Is Notepatra free? Is there a paid version?

Notepatra is free forever and licensed under GPL-3.0. There is no paid version, no subscription, no enterprise upsell, no telemetry. The source code is open on GitHub.

What makes Notepatra different from other code editors?

Notepatra is its own editor — a native Qt5 + QScintilla app with an original C++17/Rust core. It runs natively on Linux x64 / ARM64, macOS Apple Silicon, and Windows x64 from a single codebase, with a ~12 MB bare executable (~12.6 MB on Linux x64). Local-first AI is built in: pick from Ollama, llama.cpp, OpenRouter, OpenAI, Azure OpenAI, or Ollama Cloud. The Data Analyst mode works out of the box on CSV files and an in-memory SQLite engine; the Full download bundles the DuckDB v1.1.3 engine on every platform (Lite has CSV + SQLite only). The Composer tab adds a multi-file Edit Plan, `@file` mentions, Ctrl+I inline edit, AI-driven git tools (read-only: status / diff / log / branch / show), and multi-cursor editing. No Electron runtime, no telemetry, no mandatory account.

How big is the Notepatra binary?

The bare Notepatra executable is ~12 MB across every platform (12.6 MB on Linux x64). Latest v0.1.127 download sizes: 4.4 MB Linux x64 tar.gz, 4.1 MB Linux ARM64 tar.gz, 27.7 MB macOS DMG (with bundled Qt), 43.0 MB Windows MSI / 32.8 MB NSIS / 37.4 MB portable zip (with bundled Qt + QScintilla DLLs). After Windows install the on-disk footprint is ~75-85 MB because the MSI extracts the bundled Qt5 DLLs and QScintilla DLL out of the compressed payload — that's normal for any Qt app installer. Linux installs stay tiny (~12.6 MB on disk / bare) because Qt5 comes from your distro repo. Notepatra installs at ~12–85 MB depending on platform.

Does Notepatra send my code to the cloud?

No. Notepatra is local-first. The AI panel ships two local dropdown backends — Ollama and llama.cpp (llama-server, loads any GGUF directly) — and the llama.cpp entry also accepts a user-configured base URL so you can point it at any OpenAI-compatible server you have installed (examples redacted — those are separate programs that the user installs and licenses, not bundled with Notepatra). There is no telemetry, no analytics, no mandatory API key, no remote server, no phone-home in the current release. The only outbound network connections are: your chosen AI backend (when you click an AI button), git push/pull (when you click those buttons), and the REST client tab (when you explicitly send a request).

How do I install Notepatra?

On Linux or macOS run: curl -fsSL https://notepatra.org/install.sh | sh — on Windows PowerShell run: irm https://notepatra.org/install.ps1 | iex. Both scripts auto-detect your OS, download the right binary, verify the SHA-256 checksum, and install. You can also download binaries directly from the GitHub Releases page.

How do I verify the Notepatra binary I downloaded is genuine?

Every release ships with three independent integrity checks: SHA-256 checksums (sha256sum -c SHA256SUMS), Sigstore cosign signatures (keyless OIDC, recorded in the public Rekor transparency log), and SLSA build provenance attestations (verifiable with gh attestation verify). Full instructions in SECURITY.md.

Notepatra was envisioned and is maintained by Prateek Singh, with much of the implementation written with Claude's help. Single-maintainer open-source project, free forever under GPL-3.0.

Notepatra — The Code Editor Built for the AI Era

Name: Notepatra
Availability: InStock
Author: Prateek Singh

Download v0.1.127

One-command install or direct download

Auto-detecting installer scripts, or pick your platform from GitHub Releases. No telemetry. No tracking. Just a binary.

🌐

Notepatra

DEFAULT

Local AI out of the box · cloud LLMs opt-in. Talk to Ollama or llama.cpp locally by default, or any OpenAI-compatible server via the llama.cpp entry's user-configured URL. Paste an OpenAI / Anthropic / OpenRouter / Mistral / Azure URL in Settings if you want cloud — fully optional, no key required for local use.

Pick this if you want maximum flexibility — local first, with cloud available when you need a flagship model on a hard problem. Most users.

.tar.gz .deb .rpm .AppImage .dmg .msi .exe .zip

Get Notepatra v0.1.127 →

🔒

Notepatra Local AI

CLOUD-FREE

Physically cannot talk to any public LLM endpoint. api.openai.com · api.anthropic.com · openrouter.ai · api.mistral.ai · api.groq.com · everything else — refused at the network layer. Ollama / llama.cpp / LAN-hosted LLMs continue to work.

For regulated industries (finance, healthcare, legal, gov), data-sovereignty regions (EU GDPR, India RBI, China), air-gapped engineering networks. Built via -DNOTEPATRA_NO_CLOUD=ON — auditable, no setting to flip.

notepatra-local-ai_*_amd64.deb notepatra-local-ai_*_arm64.deb notepatra-local-ai-*.msi

Get Notepatra Local AI v0.1.127 →

Both flavors are local-first by default. The regular build adds optional cloud LLM URLs you paste in Settings; the Local AI build refuses them at the network layer for auditable compliance. notepatra --version tells you which flavor you have installed.

🐧

Linux x64

Binary · 4.4 MB .tar.gz

.tar.gz

🐧

Linux ARM64

aarch64 · Raspberry Pi 5 · 4.1 MB .tar.gz

.tar.gz

🍎

macOS Apple Silicon

M1 / M2 / M3 / M4 · 27.7 MB .dmg

.dmg

🪟

Windows x64

37.4 MB zip · 43.0 MB MSI · 32.8 MB setup.exe

.zip / .msi / .exe

💼 Portable .zip = most-downloaded format. Extract anywhere — USB stick, locked-down work laptop, cloud sync folder. Zero registry, zero admin. Ships with two one-click helpers: register-associations.bat adds Notepatra to Windows "Open with" for 27 file extensions (.txt .md .json .py .cpp .js .rs .go .sql .html .css and more), unregister-associations.bat undoes it. HKCU-only — never asks for admin.

AI Powered · Local-first, cloud-optional

Your AI. Your machine. Your choice.

Notepatra's AI runs locally by default — pick Ollama, llama.cpp (loads any GGUF), or any OpenAI-compatible server (any OpenAI-compatible server). When you'd rather use a frontier model, four cloud backends are wired in too: Ollama Cloud, OpenRouter, OpenAI, Azure OpenAI — six total backends in the dropdown alongside the local Ollama + llama.cpp entries. No mandatory API key. No subscription. No telemetry.

When the regex fixer can't repair your broken JSON — the AI can. Hybrid approach: speed of regex, intelligence of AI.

Explain Code Find Bugs Refactor Write Tests Fix JSON Fix HTML Fix Brackets Add Comments Generate Docs Optimize Translate Py↔JS

Supported backends — 6 in the dropdown, plus OpenAI-compat

🏠 Local (no account, no key)

• Ollama

localhost:11434 · default

• llama.cpp (GGUF)

localhost:8080 · llama-server

☁ Cloud (API key)

• OpenRouter

100+ models — Claude, GPT, Gemini, Llama, Mistral, DeepSeek, Qwen, Grok, Kimi

• Ollama Cloud

gpt-oss:120b · qwen3-coder:480b · deepseek-v3.1:671b

• OpenAI

ChatGPT API · GPT-4o, GPT-5, o-series

• Azure OpenAI

enterprise Azure resource + deployment

Plus any OpenAI-compatible server — point the llama.cpp entry at any URL/port via Settings → Preferences → AI. Notepatra speaks the standard OpenAI /v1/chat/completions protocol.

// Broken JSON input
{users: [{'id': 1, name: 'Alice',}

// Notepatra fixes it instantly:
// ✓ single quotes → double quotes
// ✓ unquoted keys quoted
// ✓ trailing commas removed
// ✓ missing braces added

{
  "users": [
    {
      "id": 1,
      "name": "Alice"
    }
  ]
}|

Features

Everything you need. Nothing you don't.

Built on QScintilla — a battle-tested Scintilla-based editor engine. Hot-path file I/O and search use Rust for memory safety on untrusted input.

🔌

MCP — your editor, readable by your AI 49 TOOLS · v0.1.127

A built-in Model Context Protocol server (notepatra-mcp) connects the assistants you already use — Claude Desktop, Claude Code, OpenAI Codex, the OpenAI Agents SDK, and any spec-compliant MCP client — to the running editor. 49 tools in three tiers: read (tabs, selection, search, Noter notes, reminders, read-only Git and SQL, diagram source, saved-connection list/query/tables), act (open, compare, format, navigate, select range, open notes, create diagram, open Noter panel, open Data Analyst, render chart), and write (edit, save, create/append notes, set reminders, replace diagram source, export diagrams, export query results, export chart) — where every write shows an Approve / Deny card inside the editor and does nothing until you click Approve (auto-deny after 120 s, no headless bypass). Local socket + stdio only — nothing leaves your machine. Read the MCP docs →

🤖

AI Assistant (Ctrl+Shift+A) NEW

Local-first AI. Two local dropdown backends: Ollama and llama.cpp (any GGUF). The llama.cpp entry accepts a user-configured URL, so it can reach anything that speaks OpenAI's /v1/chat/completions — any OpenAI-compatible server (those are separate programs you install, not bundled). Explain · Find Bugs · Refactor · Write Tests · Fix JSON · Add Comments · Optimize · Translate. No mandatory API key, no telemetry. Default is local — your code stays on your machine. Cloud LLMs (OpenAI, OpenRouter, Ollama Cloud, Azure OpenAI) are opt-in from the dropdown.

🔎

Project Search (Ctrl+Shift+G) NEW

Recursive search across file names AND file contents, project-wide. Streams line by line so memory use stays flat regardless of file size — tested on 2 GB logs. Exact line:col coordinates, double-click a match to jump the caret to the exact character. Any text-based language. Skips binary files with a NUL-byte heuristic.

🗒️

Noter — meeting notes NEW

A local-first, two-pane notes workspace — notes list on the left, editor on the right. No accounts, no bots; notes live in plain files under ~/Documents/Notepatra/Noter/. AI Extract (Ctrl+Alt+E) pulls a summary plus action items, decisions, questions, and risks. Set reminders on any note — or straight from Extract — and they're all collected in a central Reminders list (Overdue · Today · This week · Later) with desktop notifications at the due time while Notepatra is running.

📐

Diagram tool (.npd) NEW

Flow charts, ER diagrams and system designs from a tiny text DSL — the .npd source is the truth, the canvas is a live projection. Describe it in plain English and a local model writes it (review before it touches the canvas), start from a template, or type it yourself — Import Mermaid too. 5 shapes, ~55 hand-drawn icons, labelled + bidirectional arrows, 4 palettes (+ a default fallback), infinite pan/zoom. Export to PNG / JPEG / PDF on every build (plus SVG / HTML / WebP where your Qt supports them). Opens from the toolbar next to Noter. Renders natively in the default binary — every platform, no WebEngine.

🔍

AI Interaction Log NEW

Audit exactly what your AI did — transparency, not just a promise. Every request and response to any backend — local (Ollama / llama.cpp) or cloud (OpenRouter / OpenAI / Azure / Ollama Cloud), including Noter's Extract — is recorded to a local SQLite log you can open and read: timestamp, backend, model, full prompt + response, tokens, latency. Entries auto-prune after 7 days; a credential scrubber masks API keys / tokens / PEM keys before writing. The viewer has filters, Export JSON, and Prune-now. On by default with one-click opt-out in Settings. Zero network egress — purely local. Open from Features → AI Interaction Log…

⚡

Rust-Powered Core

Memory-mapped file I/O. Aho-Corasick search. Myers diff. Tested on 2 GB files. The heavy lifting runs in compiled Rust — no garbage collector, no runtime.

🔄

Safe auto-update NEW

Click Download in the in-app "update available" dialog and the right artifact for your OS + architecture is fetched, SHA-256 verified against the release's SHA256SUMS, and handed to the native installer (MSI on Windows, DMG in Finder on macOS, file manager on Linux). Your current install is never touched by the updater — the OS installer does the swap under its own transactional rollback. Checksum mismatch, cancelled download, network drop, disk full — any failure leaves the running app exactly as it was.

📅

Insert Date / Time

Edit → Insert, or Ctrl+F5. Five formats: 2026-04-20 13:45, 2026-04-20, April 20, 2026, 13:45:09, ISO-8601. Saves keystrokes in logs, changelogs, meeting notes.

🖱

Clickable status bar

Click Language, Encoding, or EOL indicators in the status bar to pop the matching change-menu right where you clicked.

📂

Clean menu organization

Three clear menus — Tools (everything built-in: AI, Terminal, Compare, JSON/HTML/SQL, Git, REST, Bracket), Plugins (third-party extensions YOU add), Utilities (Hash, Measurement). No more guessing where a feature lives.

🔀

Compare with word-level diff

Side-by-side Scintilla diff. Green for added, red for deleted, amber for changed — with LCS-based word-level intra-line highlighting so you see the exact tokens that changed. Dark mode aware. Prev/Next navigation. Ignore whitespace, case, empty lines.

📁

Session Persistence

Close Notepatra, reopen tomorrow — same files, same cursors, same window size. Crash recovery saves unsaved work every 5 seconds by default (interval configurable).

👁

File Change Detection

Someone edits your file externally? Notepatra asks: reload or keep yours. Git gutter shows changed lines in green/yellow/red.

🎙

Macro Recording

Record keystrokes (Ctrl+Shift+M), playback (Ctrl+Shift+P), run multiple times, save to .macro files, replay tomorrow.

📑

238 File Types

82 language lexers — Python, Rust, Go, C/C++, Java, JS/TS, SQL, HTML, CSS, JSON, YAML, Markdown, Bash, Fortran, Verilog, MATLAB, LaTeX and more.

🎨

3 Themes + Color Tags

Light, Dark, Monokai. Color-tag any tab (7 colors + custom). Pastel green current line, double-click word highlight in orange.

🧩

Plugin System

Drop a .so/.dll/.dylib in the plugins folder. Simple C API — export 2 functions to create a plugin. JSON, HTML, SQL, Bracket tools all built in.

🔭

Find & Replace

In-document find / replace / extended mode (\n, \t, \xNN) / regex. Ctrl+F to find, Ctrl+H to replace. For searching across multiple files, use Project Search (Ctrl+Shift+G) at the top of this grid instead.

Editor	Size	AI Built-in	JSON Fixer	Diagrams	Notes	2 GB Files	Linux	Windows	Mac	Free
Notepatra	~12 MB	✓	✓	✓	✓	✓	✓	✓	✓	✓
Notepad++	9 MB	✗	Plugin	✗	✗	✗^†	✗	✓	✗	✓
VS Code	300 MB	Extension	Extension	Extension	Extension	✗	✓	✓	✓	✓
Sublime Text	30 MB	✗	✗	✗	✗	✓	✓	✓	✓	$99
Kate	40 MB	✗	✗	✗	✗	✓	✓	✗	✗	✓
Vim	3 MB	✗	✗	Plugin	✗	✓	✓	✓	✓	✓

Security

Verifiable. Local. Hands off your data.

No analytics, no telemetry. Network traffic only on explicit actions — AI request, git push/pull, REST client, update check. Every release is checksummed and cryptographically signed. Verify before you run.

🔐

SHA-256 checksums

Every release ships a SHA256SUMS file. install.sh and install.ps1 verify it automatically and refuse to install on mismatch. Latest checksums →

✍️

Cosign signatures (Sigstore)

Every artifact is signed via keyless OIDC signing tied to the official GitHub Actions workflow. The cert is recorded in the public Rekor transparency log — no trust in any single key.

📜

SLSA build provenance

Every binary carries a cryptographic attestation linking it to the exact git commit + workflow file + runner environment that built it. Verify with gh attestation verify.

🦀

Rust safety + cargo audit

Hot-path file I/O and search are written in Rust for memory safety on untrusted input; QScintilla (C++) handles lexers and editor rendering. CI runs cargo audit on every build to flag vulnerable dependencies.

🚫

Zero telemetry

No analytics, no error reporting, no phone-home. The only outbound connections are: localhost Ollama (when you use AI), git push/pull (when you click), and HTTP requests you explicitly send via the REST tab. Verifiable with strace.

📬

Responsible disclosure

Found a vulnerability? Use private vulnerability reporting. Full policy in SECURITY.md and /.well-known/security.txt.

Version History

Releases

Every release of Notepatra, from the very first one — transparent, append-only, no surprises. Note: v0.1.4 / v0.1.6 / v0.1.11 tags exist in git but were never published as GitHub Releases — their content was rolled forward into v0.1.5, v0.1.7, and v0.1.12 respectively to reduce release churn.

v0.1.127 LATEST

2026-08-07 · Five doors refused a private key. The sixth handed it over in chunks. Retesting v0.1.126 found select_range + get_selection reading credential files — the third release running with the same defect shape. This one stops adding guards to a hand-maintained list. · Lite (default, bare) + Full · Linux · macOS (incl. Apple Silicon) · Windows

Security — the sixth door, and why there kept being another

select_range + get_selection returned credential-file contents that open_file, read_tab, find_in_tab, search_project and resources/read all correctly refused — so a private key was readable one selection at a time. get_selection had been exempted on the reasoning that a selection is human-made, so reading it back implies consent. select_range lets the client choose the range, which makes that reasoning false. apply_edit was a quieter form of the same thing: it reads the buffer as a match oracle, so "no match" versus a card naming the match reports what a file contains, one probe at a time, without returning any text.
The fix is structural, not a sixth guard. v0.1.125 unified two hand-maintained copies of the deny list; v0.1.126 added the check to four verbs; the retest found a fifth and sixth. Every fix was correct and none converged, because what kept going stale was the enumeration — every new verb had to remember to ask, and the point of a guard is that it does not depend on remembering. Buffer text now leaves the editor through exactly one accessor that refuses before returning, select_range will not even stage a selection over a credential file, and a test lints the bridge source and fails the build on any direct buffer read that bypasses the accessor.

Fixed — a marker that lied and a schema that lied

The truncation marker claimed a 5 MB cap whatever cap applied. Ask read_tab for max_bytes: 100 and the text still ended [truncated at 5 MB] — a constant from before the argument existed. It reads [truncated: showing 100 of 2688894 characters] now. The v0.1.126 notes also said read_tab "always reports truncated and total_chars"; that holds for the editor-to-sidecar wire and not for the tool result, which is plain text. The docs are corrected and the old entry left as shipped.
select_range advertised tab_id in its schema and then rejected it with -32602, so a client generating calls from the published contract failed — one verb out of thirteen slipped the v0.1.126 sweep, the one written across multiple lines. A test now walks the whole tool list and asserts every advertised tab_id is genuinely accepted. list_open_tabs also published the field as id while every verb takes tab_id; both names ship for one release.

Verification

98 C++ suites and 66 sidecar protocol tests, across both Cargo feature sets.
Every fix red-state verified by restoring the pre-fix code. The two security tests were confirmed to fail independently — the source lint flags a raw buffer read even when nobody has thought to test the verb it appears in, which is the property that makes it worth having.

Release page Full changelog Commits

v0.1.126

2026-08-07 · v0.1.126 taught search_project to skip credential files. It never told open_file or read_tab. A deep retest confirmed all four v0.1.126 Windows fixes on real hardware and found nine more defects underneath. No new editor features; every change is a fix to the MCP surface. · Lite (default, bare) + Full · Linux · macOS (incl. Apple Silicon) · Windows

Release page Full changelog Commits

v0.1.125

2026-08-07 · A one-word MCP search could return the contents of your SSH private key. Fixed, along with three other defects found by testing the MCP server on Windows — and the Windows binary finally reports its real version after 124 releases claiming 0.1.0. No new editor features; every change is a fix. · Lite (default, bare) + Full · Linux · macOS (incl. Apple Silicon) · Windows

Release page Full changelog Commits

v0.1.124

2026-07-28 · Invisible characters are now visible — 527 of them, where Notepad++ draws 113. A file containing a zero-width space used to look exactly like a file without one. View → Show Symbol is rebuilt to Notepad++'s eight-item structure, plus a toggle for the 414 codepoints its tables don't cover. Two Find/Replace bugs that were quietly destroying text are fixed. · Lite (default, bare) + Full · Linux · macOS (incl. Apple Silicon) · Windows

Release page Full changelog Commits

🎉 100 releases

v0.1.100 — one hundred shipped iterations, each one local-first, no telemetry, no account. Thank you for being here.

See every release on GitHub
Full version history from v0.1.0 onwards — release notes, signed artifacts, SHA-256 checksums, cosign signatures, SLSA build provenance.
github.com/singhpratech/notepatra/releases →

v0.2.0 — next milestone

LSP support · Plugin ecosystem · More outreach · Community feedback cycle

FAQ

Frequently asked questions

Quick answers. For deeper docs see the README and SECURITY.md.

What is Notepatra?

Notepatra is a native Qt5 + QScintilla code editor with an original C++17/Rust core for hot paths (memory-mapped I/O, Aho-Corasick search, Myers diff, formatters). Runs on Linux x64 / ARM64, macOS Apple Silicon, and Windows x64 from a single codebase. ~12 MB bare executable, no Electron runtime, no telemetry, no mandatory account. Local-first AI is built in: pick from Ollama, llama.cpp, OpenRouter, OpenAI, Azure OpenAI, or Ollama Cloud. As of v0.1.127: 238 file types · 82 language lexers, a Data Analyst mode that works out of the box on CSV files and an in-memory SQLite engine (the Full download bundles the DuckDB v1.1.3 engine on every platform), agentic Coding Mode with read / list / search / write / apply_diff tools, Cursor-style Composer + Ctrl+I inline edit, multi-cursor, AI-driven git tools (read-only: status / diff / log / branch / show).

Is it really free? What's the catch?

No catch. GPL-3.0, no telemetry, no analytics, no API key, no enterprise tier, no subscription. Source on GitHub. The "catch" is that it's a one-person open-source project — no SLA, no warranty, no support contract. See the disclaimer in the footer.

Can I install Notepatra at my company?

Yes — same license posture as Notepad++, GIMP, vim, and Linux (all GPL). Notepatra is a standalone end-user editor: under the GPL, using Notepatra as an editor does not place obligations on the files you edit (see the FSF GPL FAQ), and it doesn't phone home. For IT, legal, or procurement teams reviewing the install: see the For Enterprise page — plain-English explainer, supply-chain provenance (SHA-256 + cosign + SLSA), and a copy-paste block for your software-acceptance request.

How big is the binary?

The bare executable is ~12 MB (12.6 MB on Linux x64) across every platform. v0.1.127 download sizes: 4.4 MB Linux x64 tar.gz · 4.1 MB Linux ARM64 tar.gz · 27.7 MB macOS DMG · 43.0 MB Windows MSI / 32.8 MB NSIS / 37.4 MB portable zip. Installed on Windows the on-disk footprint grows to ~75-85 MB because the MSI extracts bundled Qt5 + QScintilla DLLs out of the compressed payload — normal for every Qt-based installer. Linux stays tiny (~12 MB on disk) because Qt is a system package; macOS and Windows bundle Qt + QScintilla DLLs for portability.

Does my code go to the cloud?

By default no — Notepatra is local-first. Pick a local AI backend (Ollama on localhost:11434 or llama.cpp's llama-server on localhost:8080) and nothing leaves your machine. The llama.cpp entry also accepts a user-configured URL, so you can route it to any OpenAI-compatible server you've installed (any OpenAI-compatible server — those are separate programs you install and license). Cloud backends (OpenRouter, OpenAI, Azure OpenAI, Ollama Cloud) are opt-in from the AI dock dropdown. Even when you opt in, the v0.1.55 credential scrubber strips API keys / tokens / private-key blocks from any code or chat content before it leaves the machine, and the privacy "Share file with AI" toggle (default OFF) controls whether the active editor file is even visible to the model. Verifiable with strace.

What languages does Notepatra highlight?

238 file types via 82 language lexers as of v0.1.127 — Python, JavaScript, TypeScript, C, C++, C#, Java, Kotlin, Rust, Go, Dart, Solidity, Zig, Vala, Hack, Julia, R, Protobuf, F#, HCL/Terraform, Thrift, GraphQL, GDScript, Nim, Cython, Mojo, Crystal, Elixir, Scala, Groovy, Apex, HTML, CSS, SQL (6 dialect presets), JSON, JSON5, YAML, TOML, Markdown, Bash, Fish, Nushell, Ruby, Perl, Lua, Fortran, Verilog, VHDL, MATLAB, LaTeX, BibTeX, Jinja, Liquid, Twig, Dockerfile, DotEnv, Gitignore, Pascal, CMake, Makefile, and more. Each lexer ships with comment / uncomment / block-comment syntax (Ctrl+Q / Ctrl+Shift+Q) and proper file-extension routing.

What file encodings are supported?

UTF-8 (with or without BOM), UTF-16 LE/BE, and UTF-32 LE/BE are auto-detected — including BOM-less UTF-16 from PowerShell and SQL Server exports, sniffed via null-byte patterns — with Latin-1 / Windows-1252 as the legacy fallback. BOMs survive a save round-trip, and the Encoding menu can reinterpret bytes or convert the save format.

How do I verify a download is genuine?

Every release ships with SHA-256 checksums, cosign signatures (Sigstore + Rekor), and SLSA build provenance. The install scripts verify SHA-256 automatically and refuse to install on mismatch. Full guide in SECURITY.md.

Will it open big files?

Yes — large files load fully. The Rust core uses memory-mapped loading (mmap2), which avoids a double read on open; the file is then held in the editor buffer (peaks at roughly 3x file size in RAM during load (~1-2x once loaded)). Files up to ~1.86 GB load fully; larger files open truncated and read-only. Files from 50 MB up to ~1.86 GB automatically disable lexer/syntax highlighting, auto-complete, word wrap, and indent guides to keep things snappy, but stay fully editable.

Can I write plugins?

Yes. Drop a .so/.dylib/.dll in ~/.config/notepatra/plugins/. The plugin C API is two functions: notepatra_plugin_name() and notepatra_plugin_run(text, len). Example in the README.

Who built it and why?

Built by Prateek Singh, constructed with Claude. The short version: Linux and macOS deserved a first-class native code editor, and 2026 deserved one with AI built in — local-first by default, cross-platform from one codebase, ~12 MB bare.

The code editor built for the AI era

Designed for people who write code.

Memory-mapped at every scale.

Private by design.

Built-in tools.

Notepatra

Notepatra Local AI

Linux x64

Linux ARM64

macOS Apple Silicon

Windows x64

MCP — your editor, readable by your AI 49 TOOLS · v0.1.127

AI Assistant (Ctrl+Shift+A) NEW

Project Search (Ctrl+Shift+G) NEW

Noter — meeting notes NEW

Diagram tool (.npd) NEW

AI Interaction Log NEW

Rust-Powered Core

Safe auto-update NEW

Insert Date / Time

Clickable status bar

Clean menu organization

Compare with word-level diff

Session Persistence

File Change Detection

Macro Recording

238 File Types

3 Themes + Color Tags

Plugin System

Find & Replace

JSON Tools

HTML Tools

Bracket Tools

SQL Formatter

Compare / Diff

Git Integration

Built-in Terminal

REST Client

Hex Editor

SHA-256 checksums

Cosign signatures (Sigstore)

SLSA build provenance

Rust safety + cargo audit

Zero telemetry

Responsible disclosure

Security — the sixth door, and why there kept being another

Fixed — a marker that lied and a schema that lied

Verification

v0.2.0 — next milestone

What is Notepatra?

Is it really free? What's the catch?

Can I install Notepatra at my company?

How big is the binary?

Does my code go to the cloud?

What languages does Notepatra highlight?

What file encodings are supported?

How do I verify a download is genuine?

Will it open big files?

Can I write plugins?

Who built it and why?

The Story

Ready to try?