Skip to content

Introduction

Katachi is a secure remote access solution for AI coding environments. It lets you access your local AI tools — Claude, Gemini, OpenAI Codex, GitHub Copilot, and custom API profiles — from any device via a feature-rich web interface.

Your code stays on your machine. Katachi creates an encrypted tunnel between your workstation and the Katachi web platform, so you can code from your phone, tablet, or any browser — without moving your files to the cloud.

flowchart LR
    A["Your Machine (Agent)"] ---|Encrypted Tunnel| B["Katachi Backend"]
    B ---|HTTPS| C["Any Browser (Web UI)"]
  1. Agent — A lightweight daemon running on your workstation. It hosts your AI providers, manages file/git operations, and establishes a secure Cloudflare tunnel.
  2. Backend — The cloud relay that authenticates users, routes requests, and manages sessions.
  3. Web UI — A full-featured IDE in your browser with file editing, Kanban-based task management, and AI chat.

A browser-based AI coding environment with a file tree, chat window, file viewer with diff support, and markdown preview — accessible from any device. You can switch the chat provider and model at any time.

Key features include a CodeMirror file viewer with syntax highlighting, file tree, tabs, and breadcrumbs; split diff view with live git status; markdown and image preview; AI-first editing via sandboxed, approval-gated tool calls; session context compaction for long conversations; and workspace memory injection (CLAUDE.md, GEMINI.md, .codex, agents, skills).

A Kanban board for managing AI-driven coding tasks. Cards move through phases — Backlog, Planning, In Progress, AI Review, Human Review, and Done. Cards support dependencies (DAG ordering), vibes mode for stricter AI review, and context injection for cross-card awareness.

An AI orchestrator that manages your Kanban board. The Session Commander plans work, creates executable cards from PR comments or issues, resolves merge conflicts, and coordinates multi-card workflows — acting as the one-stop shop for your AI coding session.

Automated code review by AI. When a task card completes its implementation, it enters AI Review where the agent reviews the diff, checks for issues, and provides feedback before moving to human review.

Switch between Claude (Agent SDK), Gemini (CLI Core), OpenAI/Codex (API), GitHub Copilot (SDK), and custom API profiles from one interface. Providers support OAuth or API key authentication (Copilot is OAuth-only).

The agent loads your workspace configuration files — CLAUDE.md, GEMINI.md, .codex, agents (.agents/agents/), skills (.agents/skills/SKILL.md), and custom commands — and injects them into the AI’s system prompt automatically.

Connect external Model Context Protocol (MCP) servers to extend the AI’s capabilities. Configure custom MCP servers per session through the MCP Registry in the web UI.

An opt-in system that gates AI tool operations in Workspace sessions. Three modes: Yolo (bypass all approvals — mandatory for Flow sessions, optional for Workspace), Default (auto-approve safe operations, prompt for destructive ones), and Strict (prompt for everything). Configure it per session in your session settings.

Full git support with worktrees, diffs, status, and branch management. Each session can operate in its own isolated worktree. The file viewer shows live git status and provides split diff views.

Quick access to actions and navigation via keyboard shortcuts.

  • Node.js 18 or later
  • Git 2.38.0 or later (required for advanced merge-tree workflows)
  • Supported Platforms: Linux (x64, arm64, arm), macOS (x64, arm64), Windows (x64)
  • A Katachi account with an active subscription
  • An AI provider subscription — Katachi does not provide AI access tokens. You need your own subscription to at least one supported provider (Claude, Gemini, OpenAI/Codex, Copilot, or Z.ai via API profile)
  • GitHub CLI (gh) — enables PR creation and issue management
  • GitLab CLI (glab) or other git provider CLIs — for PR workflows on non-GitHub repos