Stop copy-pasting between Claude, GPT, and Gemini

What it does Clowder AI sits above existing agent CLIs like Claude Code, Codex, and Gemini CLI, orchestrating them as persistent identities with shared memory and structured handoffs. It replaces the manual context-shuffling between chat windows with an A2A routing layer, cross-model review, and an evidence store that survives beyond a single session. The human stays in the loop as a “Chief Vision Officer” rather than a middle-manager.

The interesting bit The framework anthropomorphizes agents as a “clowder” of cats with self-chosen names and personalities, which is either charmingly effective or uncomfortably furry depending on your tolerance. More concretely, it bakes “Iron Laws” into both prompts and runtime code to prevent agents from deleting shared memory, killing the parent process, or touching each other’s ports.

Key highlights

• Model-agnostic adapters for Claude, GPT, Gemini, and opencode via a unified message layer.
• Built-in cross-model review workflows—Claude writes code, GPT reviews it—without human copy-paste.
• Async agent-to-agent messaging with @mention routing and thread isolation.
• Shared evidence store, decision logs, and a skills framework that loads specialized prompts on demand.
• Desktop installers for Windows and macOS; Linux currently requires source setup.

Caveats

• Google consumer Gemini CLI support ends in June 2026, so non-ACP Gemini routes already default to Antigravity CLI.
• Linux lacks a desktop installer at the moment.
• The README truncates while describing the CVO role, leaving the full feature list incomplete.

Verdict Worth a look if you’re juggling multiple model CLIs and want institutional memory without the context-window ping-pong. Skip it if you just need a single-model chat wrapper or aren’t ready to name your agents.