Why I Dropped Cursor for a 7 MB Open‑Source Terminal‑First AI IDE

What is Terax?

Terax is marketed as an AI‑native development environment (ADE), which integrates a terminal, code editor, file explorer, Git panel, web preview, and an AI sidebar into a single window.

Why It Is Small

The application’s size is only 7–8 MB because it uses Tauri 2 (Rust‑based) for the shell and React 19 for the UI, avoiding the heavyweight Electron framework used by many competing tools.

It does not collect usage data and requires no account registration, and it is released under the Apache‑2.0 license with installers for macOS, Windows, and Linux.

Terminal – Block‑Based Output and Editor‑Style Input

The terminal groups command output into blocks, making it easy to navigate history without scrolling through a continuous stream of characters. Input is provided via an editor‑style pane that supports multi‑line commands and easy corrections.

Rendering is GPU‑accelerated, so even logs with tens of thousands of lines remain smooth. It supports common shells (zsh, bash, fish) as well as PowerShell and cmd on Windows, and offers multi‑tab and split‑screen capabilities.

Two notable details: dragging a file from the file explorer or desktop into the terminal automatically converts it to a quoted, safe path; on Windows each tab can run either locally or inside a specific WSL distribution.

The “Spaces” feature restores previous tabs, working directories, and split layouts when the application is reopened.

Code Editor – AI‑Assisted Inline Completion

The editor is built on CodeMirror 6 and supports mainstream languages such as TypeScript/JavaScript, Rust, Python, and Go, with an optional Vim mode. Language servers are disabled by default and can be enabled in settings for diagnostics, go‑to‑definition, and formatting.

AI provides inline completions that can run on a local model. After the AI modifies code, changes are presented as a diff, allowing the user to accept or reject each block individually rather than overwriting the whole file.

AI Sidebar – Offline‑Ready with Custom Keys

The sidebar hosts an intelligent agent capable of reading, writing, and searching code, as well as executing bash commands after user approval. A “plan mode” lets the agent draft a plan for review before execution, and it can spawn sub‑agents for parallel tasks.

Project‑level memory is stored in a TERAX.md file at the repository root, preserving context across sessions.

The AI model layer supports providers with self‑supplied keys (e.g., DeepSeek, OpenAI‑compatible endpoints, Anthropic, Gemini) and local solutions such as Ollama, LM Studio, and MLX. Keys are stored in the system keychain, and the documentation states they never touch disk or localStorage.

Input shortcuts include # to insert saved prompt snippets and @ to attach files directly to the AI; voice input is also supported.

Claude Code can be orchestrated from the terminal, with each step requiring user confirmation.

Git Panel – Visual Commit Graph

The Git panel functions like typical graphical Git tools, allowing block‑wise staging, committing, and pushing with automatic upstream detection. The history view displays a true commit graph showing merges and branches, and commits can be searched, filtered, and opened in the remote repository.

Web Preview – Automatic Local Server Detection

When a local development server is started, the preview pane automatically detects it and opens a tab to display the site, eliminating the need to switch to an external browser. External links open in an embedded webview.

Themes – Customizable and Shareable

Users can create custom color schemes or import community themes. Background images, opacity, and blur settings are adjustable. Editor and application themes are managed separately, with built‑in options such as Kanagawa, Catppuccin, Tokyo Night, and Dracula.

Quick Start

Pre‑built installers are available on the GitHub Releases page: dmg for macOS (separate Intel and Apple Silicon packages), exe/msi for Windows, and AppImage, deb, rpm for Linux. The app checks for updates automatically.

Arch users can install via AUR with yay -S terax-bin, and Nix users can use the official flake with nix profile install github:crynta/terax-ai.

After installation, configure the AI provider and key in the settings. For local models, point the address to the Ollama or LM Studio endpoint.

To build from source, the environment needs Rust, Node 20+, and pnpm, then run:

git clone https://github.com/crynta/terax-ai.git
cd terax-ai
pnpm install
pnpm tauri dev

Packaging is done with pnpm tauri build.

Who Should Use It?

Front‑end developers benefit from the integrated terminal, editor, and live preview, enabling a seamless workflow without switching windows.

Heavy terminal users (e.g., tmux enthusiasts) gain a fully equipped environment plus an AI assistant.

Security‑sensitive environments can run local models (e.g., Ollama) to keep code and data on‑premise.

Existing Claude Code users can use Terax as a unified orchestration hub.

Author’s Opinion

The author appreciates the tool’s restraint: no registration, a tiny 7 MB footprint, no data collection, and the ability to use personal API keys or local models. Although the project is young (open‑sourced in April) and may have minor rough edges, the frequent commits and active issue/PR handling suggest rapid development. The author recommends trying it, especially for terminal‑centric workflows.

Open‑Source Repository

https://github.com/crynta/terax-ai