7/24 IDE

Always On. Always Coding.

An autonomous, local AI developer agent for Windows.
Describe what you want — it reads your files, writes precise edits, runs commands, and shows you a live preview, all inside your own project folder.

Why 7/24 IDE?

Standard AI chats make you copy-paste code blocks endlessly. Browser sandboxes trap you in tiny web demos with no way to install native packages or run real backends.

7/24 IDE is different. It's a focused, local workspace built around an autonomous developer agent. You describe a task; the agent reads your files, makes precise modifications, runs install/build scripts, and renders a live preview — like a junior developer working directly in your project folder, 24/7.

Everything runs locally. Your code, chats, settings and snapshots never leave your machine.

Plan & Build — two ways to work

Mode	Best for	How it works
⚡ Build	Fixes, refactors, incremental changes	The agent edits files and runs commands immediately. A live activity bar shows what it's doing, which files changed, and token usage.
🗺️ Plan	Whole features from scratch	The agent drafts a step-by-step plan you approve. Each step runs in an isolated micro-agent inside a shadow workspace, merged only on success, with a sticky progress bar and self-healing on build errors.

Key Features

⚡ Native Rust Engine

An optional bundled Rust core builds real Tree-sitter syntax trees (Rust, TypeScript/TSX, JavaScript/JSX, Python, HTML, CSS, JSON) and ranks workspace code search with BM25 relevance. The workspace is indexed in the background. If the binary isn't present, the app falls back transparently to its TypeScript implementation — no loss of functionality.

🤖 Smart Model Picker

Search hundreds of OpenRouter models in a clean dropdown with FREE badges, context-window and price labels, and tabs for All / Free / ★ Favorites. Pin favorites with a single star. Configure a fallback model for transient provider failures.

🦙 Ollama Offline Mode

Run fully private, offline builds. Select the Ollama provider, point it at your local endpoint (http://localhost:11434), fetch local models and tune the context window (num_ctx) — no cloud, no API key.

🛠️ Model Context Protocol (MCP)

Extend the agent with external stdio JSON-RPC MCP servers added straight from settings — database inspectors, search engines, web readers, formatters and more.

🛡️ Safety & Absolute Control

Auto-Checkpoints — a silent snapshot is taken before every agent run; roll the whole project back in one click from the Snapshots tab.
Monaco Diff Reviews — inspect every change side-by-side in a fullscreen diff before it touches disk.
Permissions Panel — per-operation rules (Always Allow / Ask / Deny) for file reads, writes and command execution.
Folder-Scoped Sandbox — the agent is locked inside your chosen workspace folder.

💬 2026-Ready Reasoning Chat

Collapsible reasoning blocks, clean tool-call cards, anti-truncation auto-continue, copy & Markdown export, and a readable centered chat column. Tool parsing tolerates model quote-style variations so actions keep running instead of leaking raw XML.

🖼️ Image Paste, Branching & Run-from-Chat

Paste (Ctrl+V) or drag-and-drop images as context. Branch any conversation to experiment without losing history. Run JavaScript/TypeScript/Python/Shell code blocks directly from chat with one click.

🤖 Smart Git Auto-Commits

Optionally stage and commit after each plan step with AI-generated messages. Verify/edit the message before it lands, and customize the prefix (defaults to [AI]).

🌍 Localized Interface

Full UI available in English, Russian and Chinese.

⌨️ Keyboard Shortcuts

Shortcut	Action
`Ctrl+K`	Quick chat search
`Ctrl+L`	Clear chat / new
`Ctrl+Shift+M`	Toggle Build / Plan
`Ctrl+N`	New project
`Ctrl+,`	Open Settings

Quick Start

Download & Install the latest installer from the Releases page (7-24-IDE-Setup-x.y.z.exe), or grab the portable build.
Pick a Workspace — choose any folder on your computer where you want to build.
Connect a Model — paste an OpenRouter API key for cloud models (Claude, GPT-4o, Gemini, Llama, etc.), or point to a local Ollama endpoint for fully offline work.
Start Building — describe what you want (e.g. "Build a weather dashboard with city search"), pick Build or Plan, and watch it come together.

Your API key is encrypted with the OS keychain (Windows DPAPI). All data lives locally in %APPDATA%\7-24 IDE\. No telemetry, no cloud storage, no lock-in.

Under the Hood

Native Rust Core (core-backend): spawned on startup over JSON-RPC stdio for Tree-sitter AST parsing and BM25 code search. Build it with cargo build --release in core-backend/ — or just npm run build in desktop-gui/, which auto-detects cargo and bundles the binary as extraResources. Skipped gracefully if cargo is absent.
Dual-mode workspace: Build (direct edits) and Plan (step-by-step checklist executed by isolated micro-agents in a shadow workspace).
AST compression: large files are folded to signatures so they don't consume the whole context budget.
Self-evolving skills: after a successful build the agent reflects on the session and compiles reusable rule sets that auto-activate on similar tasks.
Atomic writes: files are written to a temp sibling and renamed, preventing corruption on interruption.
Prompt caching: optimized system prompts for Anthropic models lower latency and cost.

Building from Source

# Desktop app (Electron + TypeScript)
cd desktop-gui
npm install
npm run build      # bundles main/preload/renderer + Monaco (+ Rust core if cargo present)
npm start          # launch the app
npm test           # run the test suite
npm run dist       # build the Windows NSIS installer into ../dist-installer

# Native engine (optional, requires Rust toolchain)
cd core-backend
cargo build --release

License

Released under the MIT License.