Minimal, private AI-native terminal

---

Kai is a fast, cross-platform terminal built on Tauri 2 + Rust and React 19. Terminal, code editor, file explorer, web preview, and AI agent — in one app, lightweight, zero telemetry.

Contents

• Features · Agent Personas · Install · Quick Start · Keyboard Shortcuts · Configuration Guides · Build from Source · Tech Stack

Features

Terminal

• xterm.js + WebGL rendering, multi-tab with split panes
• Shell integration for bash, zsh, PowerShell, and cmd
• Inline search, smart error detection, autocomplete
• Escape-key streaming interrupt, and a toroidal Conway's Game of Life thinking spinner

Editor

• CodeMirror 6 with 44 languages and find & replace
• Side-by-side AI edit diffs with per-hunk approve/reject
• 9 editor themes, Vim mode (:w / :q Ex commands, visual/insert/normal)
• Image & PDF preview, plus Open in Live Preview — right-click an HTML file to serve and preview it in-app

AI Agent

• Bring your own key — 12 providers: OpenAI, Anthropic, Google, Groq, xAI, Cerebras, DeepSeek, Mistral, OpenRouter, and direct Z.ai (GLM), plus any OpenAI-compatible endpoint
• Local models via LM Studio or Ollama — vision models receive images automatically (runtime-name detection + a manual Vision override in Settings)
• Paste screenshots straight into chat with Ctrl+V; attach PDF/DOCX via the + button (text extracted, works on every model)
• /compact — reset the context in place in the same chat when a session runs long; no new chat needed
• Screen access — look_at_screen shows your displays to the agent (multi-monitor aware, every capture needs your approval)
• Universal file extraction — zip, audio, image, and video read locally as text
• Interactive todo toggling with auto-closing completion lists
• Per-model thinking control — a toggle on each reasoning model cycles effort off/low/medium/high
• Multi-step goals with automation, plan mode (queue edits for review as one diff), and sub-agents (explore / code-review / security / research)
• Conversation forking and merging — branch any message into a parallel thread, then merge it back or into another thread

Image & Video Generation

• Generate directly from chat, with inline rendering, lightbox, and download
• Image: OpenAI GPT Image 2 · Google Nano Banana 2 · xAI Grok Imagine · ComfyUI
• Video: Kling 3.0 · Google Veo 3.1 · ByteDance Seedance 2.0 · ComfyUI

Git

• Stage, commit, push, and review diffs from the source-control panel
• Manage branches (list, switch, create, delete) without leaving the panel — new branches publish with one click (push -u sets up tracking)
• Open the GitHub/GitLab/Bitbucket pull-request page for the current branch, straight from the panel
• Resolve merge conflicts with conflict counts and per-file line locations, plus one-click "Ask KAI to resolve"
• Optional GPG commit signing (auto-sign or approval modes)

Extensibility & Tooling

• MCP — connect external tool servers and install from the official registry, or hand the agent a setup URL: mcp_add_server registers it so it shows up live in the MCP Servers UI
• File explorer and built-in web preview (with one-tap Stop)
• REST API tester and 8 UI themes
• Voice input — OpenAI whisper-1 transcription with a no-key browser SpeechRecognition fallback
• PDF/DOCX reading, image OCR (tesseract), ZIP/JAR listing, and audio metadata
• Direct PDF export (convert_to_pdf) from Markdown/Word
• YouTube transcript summarization
• Kai.md project memory, context summarization, and auto-approve modes — plus save/red persistent memory tools the agent can use mid-session
• Skills & snippets
• Per-project sandbox (read-only or workspace-only) and a detached-copy workflow — work in an isolated copy, then merge or discard

Agent Personas

KAI ships highly specialized built-in agent personas for different kinds of work. Switch between them in the chat pane, customize their prompts, and reset to factory defaults at any time.

• Coder 💻 — general-purpose development: resolves, refactors, and runs tests
• Architect 📐 — tradeoffs, scalability, and multi-option blueprints
• Code Reviewer 🔍 — logic, edge cases, race conditions, and performance cliffs
• Security 🛡️ — threat modeling, input validation, and secure crypto defaults
• Designer 🎨 — UI/UX critique, density, typography, spacing, and modern motion
• Researcher 🛰️ — web search, page browsing, and structured citations
• Assistant ✍️ — brainstorming, copyediting, summaries, and text tasks

Install

Grab the latest installer from Releases — available for Windows (.exe), macOS (.dmg), and Linux (.deb / .rpm / .AppImage).

Auto-update is built in.

Quick Start

1. Open Settings → Models
2. Add an API key for any provider — or point to a local model
3. Press Ctrl+I to open the AI agent

Keys are stored in the OS keychain. No account required.

Keyboard Shortcuts

Shortcuts use ⌘ on macOS and Ctrl on Windows/Linux.

Action	Shortcut
Toggle AI agent	⌘ I
Ask AI about selection	⌘ L
Toggle voice input	⌘ M
New tab · private terminal	⌘ T · ⌘ R
New editor · preview tab	⌘ E · ⌘ P
Close tab / pane	⌘ W
Split pane right · down	⌘ D · ⌘ ⇧ D
Next · previous pane	⌘ ] · ⌘ [
Search files	⌘ ⇧ F
Find · find & replace	⌘ F · ⌘ H
Toggle file explorer	⌘ B
Toggle source panel	⌘ G
Show shortcuts · settings	⌘ K · ⌘ ,
Zoom in · out · reset	⌘ = · ⌘ - · ⌘ 0

Configuration Guides

1. Offline Setup with LM Studio

KAI has full support for local, offline-only development. To use a local GGUF model:

1. Open LM Studio and go to the Developer tab.
2. Start the local HTTP server and note the base URL (usually http://localhost:1234/v1).
3. In KAI, open Settings → Models and enter the base URL and active Model ID (e.g. qwen3.3-coder-instruct). Click Save.
4. Select LM Studio (Local) in chat for fully private intelligence.

2. One-Click MCP installations

Extend your AI agent's capabilities with custom tools via the Model Context Protocol:

1. Open the left sidebar and switch to the Extensions (puzzle icon) tab.
2. Browse or search for official MCP servers (e.g. Filesystem, PostgreSQL, or GitHub).
3. Click Install — KAI downloads and connects the server in the background, making its tools immediately available to the active agent.

3. Project Memory with Kai.md

Define custom guidelines, development standards, architectural context, and preferred workflows on a per-project basis. Create a file named Kai.md at the project root and KAI automatically injects its contents (up to 32KB) into the agent's instruction prompt on every message.

• Case sensitivity — the filename must be exactly Kai.md on case-sensitive filesystems (Linux). On Windows/macOS, KAI.md / kai.md are also recognized.
• Zero-write safe — KAI only reads this file; it never writes to it.
• What to include — tech stack, formatting guidelines, database schemas, directory mappings, or preferred test commands.

4. Anthropic models on a Claude subscription (in the terminal)

KAI's built-in AI agent is BYOK — Anthropic models selected in Settings → Models use your Anthropic API key and bill at standard API rates. If you have a Claude Pro/Max subscription and want to use it instead, run Anthropic's official Claude Code CLI in a KAI terminal pane. KAI is a full terminal, so the CLI behaves exactly as it would anywhere else and authenticates against your subscription.

1. Install Claude Code — npm install -g @anthropic-ai/claude-code (or the native installer) — then verify with claude --version.
2. cd into your project, run claude, and log in with your Claude account (Pro/Max) — not an API key. Switch later with /login, and confirm with /status.
3. Make sure ANTHROPIC_API_KEY is not set in the shell — if it is, Claude Code may bill the API instead of your subscription. Check with echo $ANTHROPIC_API_KEY.
4. Give Claude Code its own terminal tab. KAI keeps background PTYs alive across tab switches, so the session keeps running while you work in the editor.

Why the terminal: per Anthropic's terms, Claude subscription auth is only for its official clients (Claude Code, claude.ai). Running the official CLI in a terminal is the supported way to use your subscription; KAI's own AI agent remains API-key based and unaffected.

Build from Source

# Prerequisites: Rust (stable), Node 20+, pnpm, Tauri v2 deps
pnpm install
pnpm tauri dev          # development
pnpm tauri build        # production

Tech Stack

Tauri 2 · Rust · React 19 · TypeScript · xterm.js · CodeMirror 6 · Vercel AI SDK · Tailwind v4 · shadcn/ui

Contributing

Issues and PRs welcome.

License

Apache-2.0 — see LICENSE and NOTICE.