Philosophy

Nexis is shaped by a few firm principles. They explain most of the product decisions you’ll run into.

Local-first and private

Your code and your keys stay on your machine unless you choose a networked feature. API keys are saved through the OS keychain via Rust’s keyring crate, not in Nexis settings or localStorage. The OS may persist its keychain on disk. Nexis does not include usage telemetry.

If you don’t want to use a cloud model at all, you don’t have to — Nexis runs fully offline against LM Studio, MLX, or Ollama.

Native by construction

Nexis is built on Tauri 2 rather than Electron. It renders in the system webview and does the heavy lifting in Rust. The project no longer treats a tiny binary as a hard constraint: useful native capabilities such as bundled CPU-only ONNX Runtime take precedence over an arbitrary size target.

AI with a seatbelt

The AI is powerful but governed by visible policies:

• Tool policies are configurable per capability: prompt every time, deny, allow, or auto-approve only shell commands that pass a strict read-only check.
• AI-proposed edits appear as per-hunk diffs — approve or reject each change.
• The AI context inspector shows exactly what is being sent to the model.
• Private terminals are walled off: the AI cannot read their scrollback.

Transparent and open

Nexis is open source under the Apache-2.0 license, and builds on the open-source terax-ai project. The wire protocol between the app and its ML engines is a documented NDJSON format, and the ML run store is a plain on-disk directory you can inspect.

One window, not five

A terminal, an editor, a debugger, source control, and an AI assistant belong in the same window with shared context. Nexis puts them there so you’re not juggling an editor, a separate terminal, and a chat window that each know nothing about the others.