Kavi

Kavi

The multi-agent TUI orchestrator for your terminal.

Kavi puts Claude Code and OpenAI Codex side-by-side in a single terminal session,
routes work to the right agent, and merges their output into one clean result.

[](https://www.npmjs.com/package/@mandipadk7/kavi) [](./LICENSE) [](https://nodejs.org)

[Install](#install)  ·  [Quick Start](#quick-start)  ·  [Features](#features)  ·  [Commands](#commands)  ·  [Configuration](#configuration)  ·  [Architecture](#architecture)  ·  [Contributing](#contributing)

Why Kavi?

You already use Claude Code. Maybe you use Codex too. But switching between them is friction — separate terminals, separate context, no coordination.

Kavi fixes that. It runs both agents as headless child processes, gives each its own git worktree, routes tasks based on what each agent is best at, and merges their changes when they're done. You interact with Kavi; Kavi talks to the agents.

**What you get:**

• One terminal, two agents. Claude and Codex run simultaneously in isolated worktrees. No tab-switching, no copy-pasting context.
• Smart routing. Kavi decides which agent handles each task — by keyword rules, path ownership, or AI classification. Override anytime.
• DAG-based planning. Complex goals get decomposed into a task graph with dependencies. Parallel where possible, sequential where required.
• Approval gates. Every tool invocation (shell commands, file writes) goes t