Ensemble

Multi-agent collaboration engine — AI agents that work as one.

Ensemble orchestrates AI agents into collaborative teams. Out of the box it pairs Codex (lead) + Claude Code (worker), and they communicate, share findings, and solve problems together in real time. Teams of three (adding Grok) work too.

Status: Experimental developer tool. Not a production framework (yet).

Default team: Codex (lead) + Claude Code (worker). You can add other agents like Grok, GLM, Antigravity CLI (agy) or any custom CLI tool, and run teams of three.

What does it do?

You give a task. Ensemble spawns a team of AI agents, each in their own tmux session, that talk to each other to solve it. You watch the conversation unfold in real time via a TUI monitor or inline feed.

You: "Review the auth module for security issues"

  codex-1: I'll audit the config, entitlements, and privacy settings.
  claude-2: Got it. I'll focus on the Swift code, crash risks, dead code, performance.
  grok-3:  Then I'll take the network layer and cross-check what you two find.
  codex-1: Found hardcoded API key in NetworkService.swift line 42...
  claude-2: Confirmed. Also found unvalidated user input in AuthHandler.swift...

Which agents can join

Any CLI-based coding agent can be a team member. These ship in agents.json:

Agent Status Notes
Codex Fully tested, default lead OpenAI Codex CLI
Claude Code Fully tested, default worker Anthropic Claude Code
Grok Tested in three-agent teams Needs the project-picker hint, see Configuration
GLM Tested in four-agent teams Claude Code pointed at the Z.ai endpoint, so a GLM model joins as a regular agent
Antigravity CLI (agy) Configured Uses subscription authentication
MiMo Code CLI (mimo) Configured Uses file-based prompt delivery
Gemini CLI Legacy Configuration retained
opencode Untested Config included, not battle-tested
Any CLI tool Custom Add your own

Pick them per run, or set a line-up once:

./scripts/collab-launch.sh "$(pwd)" "Security audit" codex,claude,grok
export COLLAB_AGENTS="codex,claude,grok"    # same thing, every run

Use with Claude Code

Ensemble ships with a /collab skill for Claude Code. Just type:

/collab "Review the auth module for security issues"

Claude spawns a team, shows the agent conversation live, and presents results when they’re done. Setup in one command:

./scripts/setup-claude-code.sh

See Configuration → Claude Code integration for details.

Key features

  • Team orchestration — Spawn multi-agent teams with a single command
  • Real-time messaging — Agents communicate via a structured message bus
  • TUI monitor — Watch agent collaboration live from your terminal
  • Extensible — Add any CLI-based AI agent via agents.json
  • Multi-host support — Run agents across local and remote machines
  • Auto-disband — Intelligent completion detection ends teams when work is done
  • Telegram notifications — Get notified when teams finish