Crynta released Terax, a lightweight open-source terminal-first AI-native development workspace, on GitHub. At 7 to 8 MB on disk, it is a fraction of the size of VS Code (which ships at roughly 300 MB) or any Electron-based IDE. The project uses Tauri 2 for the shell, Rust for the native PTY backend, and React 19 for the frontend. It bundles a multi-tab terminal, a CodeMirror 6 code editor, a source control panel with a git graph, a file explorer, a web preview pane, and an agentic AI side-panel. No telemetry. No account required.

The AI side-panel is the headline feature. It supports BYOK providers: OpenAI, Anthropic, Google (Gemini), Groq, xAI (Grok), Cerebras, OpenRouter, DeepSeek, Mistral, and any OpenAI-compatible endpoint. For local inference, it connects to LM Studio, MLX, or Ollama. The agentic workflow includes plans, sub-agents, project memory via a TERAX.md file, and a tool set covering file read, write, edit, multi-edit, grep, glob, and bash execution with approval gating. Keys are stored in the OS keychain via the keyring crate, never on disk or in localStorage.

The size matters. A 7 MB AI-native dev workspace that runs local models and does not phone home is a direct rebuttal to the assumption that AI tooling must be a cloud-dependent, resource-hungry Electron app. VS Code with the GitHub Copilot extension ships telemetry by default and requires an internet connection for completions. Cursor, Windsurf, and other AI-first IDEs are built on Electron, typically consume hundreds of megabytes, and rely on proprietary backend services. Terax proves that a developer can get a comparable experience with no cloud dependency, no tracking, and a disk footprint that fits on a floppy disk.

The terminal is the center of gravity. Terax uses xterm.js with a WebGL renderer, multi-tab support with background streaming, and a native PTY backend via portable-pty. It supports zsh, bash, pwsh, fish, and cmd. Split panels work horizontally and vertically. The terminal is GPU-accelerated and includes inline search, link detection, and true-color support. This is not a terminal bolted onto an editor. It is a terminal that happens to have an editor and AI attached. That design choice reflects a real workflow: many developers spend more time in the terminal than in the editor, and the AI should be available wherever the developer is working.

The code editor is CodeMirror 6, which supports all popular languages. It includes inline AI autocomplete with local model support, AI edit diffs with hunk-by-hunk accept or reject, and Vim mode. Ten built-in editor themes ship with the app. The editor is not the star, which is the point. Terax treats the editor as one pane among several, not the organizing principle of the workspace.

The source control panel includes staging and unstaging hunks, committing with Cmd+Enter, pushing with upstream awareness, branch display including detached HEAD state, a git history pane with lane rendering for merges and branches, commit search and filter, and click-through to the remote commit page. The file explorer uses the Catppuccin icon theme, supports fuzzy search, keyboard navigation, inline rename, and context actions. Files and selections can be attached directly to the AI side-panel.

The web preview auto-detects local dev servers and opens them in a preview tab. External URLs open via a native child webview. The app supports custom themes built in-app, background images with adjustable opacity and blur, and independent editor and app themes.

Terax is licensed under Apache 2.0. The build process requires Rust, Node 20+, pnpm, and Tauri prerequisites. The tech stack is Tauri 2, Rust, portable-pty, React 19, TypeScript, Vite, xterm.js, CodeMirror 6, the Vercel AI SDK v6, Tailwind v4, shadcn/ui, and Zustand. Installers are available for Windows, Linux (including Arch AUR and NixOS flakes), and AppImage. The project is not code-signed yet, so Windows shows a SmartScreen warning on first launch.

The project’s GitHub star history shows growing interest. The repository is active, with issues and PRs welcome. The documentation is minimal but functional. The website and docs are also open-source.

What Terax implies for the AI tooling landscape is straightforward. The market for AI-native development environments is splitting. One path is the heavy, cloud-dependent, proprietary IDE (Cursor, Windsurf, GitHub Copilot in VS Code). The other path is the lightweight, local-first, open-source workspace (Terax, and to some extent Zed, which is also Rust-based but heavier at around 50 MB and focused on collaborative editing rather than terminal-first workflows). Terax is the most extreme version of the second path: terminal-first, local-first, no telemetry, no account, 7 MB.

The tradeoff is real. Terax does not have the deep editor integrations that Cursor offers. It does not have the collaborative features of Zed. It does not have the extension ecosystem of VS Code. What it has is a clear philosophy: the developer should own their tools, the AI should run locally if desired, and the terminal should be the center of the workspace. For developers who work primarily in the terminal, who care about privacy, who want to run local models, and who are tired of Electron bloat, Terax is a compelling alternative.

The open-source license is significant. Cursor and Windsurf are proprietary. GitHub Copilot is proprietary and tied to a subscription. Terax is Apache 2.0. That means any developer can fork it, audit it, modify it, and redistribute it. The project’s use of the Vercel AI SDK v6 is notable because that SDK is also open-source and supports multiple providers, including local ones. The combination of an open-source IDE and an open-source AI SDK creates a stack that no single vendor controls.

The project’s approach to agentic workflows is pragmatic. Plans, sub-agents, project memory, tool gating. These are not research contributions. They are engineering decisions about how to make an AI assistant useful in a terminal context. The TERAX.md file for project memory is a simple convention that any developer can understand and modify. The bash approval gating means the developer stays in control of what commands the AI runs.

Terax is not a finished product. It is a 7 MB open-source project that ships installers and auto-updates. It has rough edges: no code signing on Windows, Wayland rendering glitches on Linux, no extension API, no debugger. The documentation is sparse. The community is small. But the core idea is sound, and the implementation is clean.

The question for AI builders is whether the terminal-first, local-first, open-source approach scales. Can a 7 MB IDE built on Tauri and Rust compete with the resources Microsoft and the VC-backed startups are pouring into AI dev tools? The answer depends on what developers value. If they value speed, privacy, and ownership, Terax has a path. If they value deep editor integrations, collaborative features, and a polished out-of-the-box experience, the heavy IDEs will win. The market is not zero-sum. Both approaches will coexist.

Terax ships at 7 MB and does not ask for an account. That is not a feature. It is a statement.