This is an early release preview. You may encounter bugs.
Claude Codex Bridge logo
Unclaimed

Tool coding interface

Claude Codex Bridge

Visible multi-agent CLI workspace for mixing Codex, Claude, Gemini, Kimi, Qwen, Cursor, Copilot, Pi, OpenCode, and other AI coding agents

B 73/100 GitHub score ? This grade is derived from GitHub signals, not user votes. Open for the full breakdown.
Discussion

01 / Summary

About Claude Codex Bridge.

Claude Codex Bridge (CCB) is a terminal workspace that runs several CLI coding agents side by side in managed tmux panes and lets them hand work to one another. Each interactive agent (Codex, Claude, Gemini, Pi, OpenCode, Kimi, Qwen, Cursor, GitHub Copilot, Grok CLI, and others, 16 CLI families in all) keeps its full native terminal, so you can watch what it does and take over by typing in its pane, while a background daemon keeps project state alive when the foreground UI is closed.

Collaboration runs through an /ask command: you or an agent addresses another agent by name, and that agent can delegate further, which supports collaboration graphs such as one agent feeding two or two agents feeding one. Requests and returned results share a chronological queue per target agent, so a later reply cannot interrupt an earlier reply's turn, and incoming agent messages wait while you compose a draft in that pane. A project-wide shared memory file holds collaboration rules, constraints, and handoff conventions for every agent.

A workspace starts with one window and a single agent that uses the first supported CLI found on the machine. A local configuration control panel edits windows, pane splits, providers, models, thinking levels, API overrides, and workspaces, validates changes, and supports guarded hot reload, and a config file describes multi-agent topologies such as a main agent, two workers in git worktrees, and a reviewer and QA pair. CCB supports the Agent Roles Spec for packaging specialist agents as installable Role Packs that bundle skills, memory, and tool dependencies, with a public catalog of roles for architecture review, frontend, mobile, and workflow orchestration.

An Android app acts as a remote controller: it discovers projects through a server-wide mobile gateway, switches windows and agents, sends text into panes, opens a terminal view, accepts voice input, and transfers files, with remote access over Tailscale Serve and scopes limited by the pairing profile. An optional rich-media mode opens a WezTerm launcher for browsing file trees, editing documents, and previewing media inside the terminal. CCB runs on Linux, macOS, and WSL, with a native Windows x64 beta.

  • Multi-agent panes: Codex, Claude, Gemini, Pi, OpenCode, Kimi, Qwen, Cursor, GitHub Copilot, Grok CLI, and other CLI agents run concurrently as native terminals you can take over
  • Agent-to-agent asks: /ask delegates work between agents and supports graphs such as A -> B -> C, A,B -> C, and A -> B,C
  • Message queues: requests and returned results share one FIFO queue per agent, and deliveries wait while a human draft sits in the input box
  • Background daemon: project state survives closing the foreground UI
  • Configuration control panel: edits windows, splits, providers, models, thinking levels, and API overrides with validation, dry-run reload, and guarded hot reload
  • Role Packs: Agent Roles Spec bundles of skills, memory, and tool dependencies, installed from a public role catalog
  • Shared memory: a project-wide memory document for team rules, constraints, and handoff conventions
  • Mobile remote control: an Android app with agent chat, terminal view, voice input, and file transfer over an authenticated loopback, LAN, or Tailscale Serve gateway
  • Rich media terminal: a WezTerm-based mode for file browsing, document editing, and media preview
  • Control skills: built-in ask, clear, compact, and diagnose skills for every managed agent, including runtime diagnosis with bounded recovery
  • Transactional updates: the updater rolls back on failure, pins versions, and offers provider CLI updates

02 / Discussion CREDIBILITY-GATED

Discussion

Reading is open to everyone. Posting and voting need a verified identity or a GitHub grade of B or higher.

  • No discussions yet.

04 / Build

Build with Claude Codex Bridge.

Browse the catalog for frameworks, tools, and harnesses, each scored on real GitHub credibility.

Get Claude Codex Bridge →

Browse the catalog