What does /mission actually do?

You give Medley a prompt — like "get my first 10 customers." /mission turns it into a full multi-agent plan: the target, the budget, the deadline, and what to check with you first. Then it runs a coordinated team of agents on the steps.

Is Medley a replacement for Claude Code?

No. Medley is a plugin that makes the harness you already use stronger — Claude Code or Codex. You keep working where you work; Medley adds multi-agent decomposition, parallel execution, and a URL to see the structure.

Why does /mission beat a single model?

Because it runs a coordinated multi-agent plan instead of one model doing everything alone. That can mean parallel agents, different models on different steps, or both — multi-agent beats a single model alone.

Does Medley act without my permission?

No. Agents run inside the guardrails you set, and Medley asks before anything sensitive — publishing, sending outbound, spending, or contacting people. You grant more autonomy at your own pace.

Medley is free. It uses the models you already have access to in Claude Code or Codex. Install the plugin and start a mission.

Backed by Y Combinator

Start a mission. Get a coordinated team of agents on the work.

Medley is a plugin for Claude Code and Codex. It decomposes your prompt into a multi-agent plan and gives you a URL to watch the structure. Multi-agent beats a single model alone.

Add the Medley marketplace
/plugin marketplace add Spine-AI/medley
Install the plugin
/plugin install medley
Reload plugins
/reload-plugins

Then start a mission: /mission <your goal>

Check your Codex version (needs 0.142.0+)
codex --version
Add the Medley marketplace
codex plugin marketplace add https://github.com/Spine-AI/medley
Install the plugin
codex plugin add medley@medley

Then start a mission: $medley:mission <your goal>

One prompt. A coordinated plan that delivers.

Type /mission with a prompt. Medley breaks the work into a visible plan and runs a coordinated team of agents on it — sequentially, in parallel, or overnight. Multi-agent beats a single model alone.

$ /mission Find a GTM channel with real signal

Mission: GTM channel test
├─ Task 1: Research channels → Claude
├─ Task 2: Draft outreach → Claude
├─ Task 3: Build lead list → GPT
└─ Task 4: Review before send → you

A whole team of agents. Inside Claude Code or Codex.

Every task in a mission runs on one of these — Claude Code, Codex, Cursor, OpenCode, Kimi, or Pi. Medley routes each step to the agent that fits and uses whichever you already have signed in, so you get better results than one model working alone.

Claude Code
Codex
Cursor
OpenCode
Kimi
Pi

Go farther. Do more.

Medley missions reach beyond a single coding session — research, outreach, docs, calendars, and the apps your work already lives in. Structure complex work that a single coding session was never going to finish.

Lives inside Claude Code and Codex.

No separate app to babysit. Start a mission in the harness you already use — Medley decomposes the work, runs a coordinated multi-agent plan, and gives you a URL to see the structure.

Just type /mission

Invoke Medley with a prompt — /mission in Claude Code, $medley:mission on Codex. Decomposition and multi-agent execution start from the terminal you already live in.

Many agents, one mission

A coordinated team of agents on the work — not one lonely model. Multi-agent beats a single model alone.

A URL for the structure

Open the mission board in a browser to see the same task tree, status, and handoffs you see in the terminal.

Try a /mission

Pick a use case and see exactly how Medley works.

admin — ✱ Claude Code — medley — 120×30

Claude Code v2.1.209

Opus 4.8 (1M context) · Claude Team

~/medley

Built by builders from

Benchmarks

Better than any single model — because multi-agent wins.

Across four official benchmarks, Medley leads the highest reported Anthropic and OpenAI family scores. Multi-agent beats a single model alone.

93.25%

Terminal-Bench 2.1Task pass rate

+1.35 pts vs. prior best

71.37

HealthBench ProfessionalMean rubric score

+5.37 pts vs. prior best

93.24%

ViBenchPlan Pass@1

+5.44 pts vs. prior best

60.98%

DrugDiscoveryBenchExact-answer pass rate

+9.38 pts vs. prior best

Benchmark	Medley	Anthropic	OpenAI	Lead
Terminal-Bench 2.1Task pass rate	93.25%	88.0%	91.9%	+1.35
HealthBench ProfessionalMean rubric score	71.37	66.0	60.5	+5.37
ViBenchPlan Pass@1	93.24%	87.8%	86.5%	+5.44
DrugDiscoveryBenchExact-answer pass rate	60.98%	50.1%	51.6%	+9.38

Compare with

Questions

What does /mission actually do?: You give Medley a prompt — like "get my first 10 customers." /mission turns it into a full multi-agent plan: the target, the budget, the deadline, and what to check with you first. Then it runs a coordinated team of agents on the steps.
Is Medley a replacement for Claude Code?: No. Medley is a plugin that makes the harness you already use stronger — Claude Code or Codex. You keep working where you work; Medley adds multi-agent decomposition, parallel execution, and a URL to see the structure.
Why does /mission beat a single model?: Because it runs a coordinated multi-agent plan instead of one model doing everything alone. That can mean parallel agents, different models on different steps, or both — multi-agent beats a single model alone.
Does Medley act without my permission?: No. Agents run inside the guardrails you set, and Medley asks before anything sensitive — publishing, sending outbound, spending, or contacting people. You grant more autonomy at your own pace.
What does it cost?: Medley is free. It uses the models you already have access to in Claude Code or Codex. Install the plugin and start a mission.

See all questions →

Install /mission in three commands.

Add the Medley plugin to Claude Code or Codex, start a mission, and get a coordinated team on the work — with a URL to watch the structure.

Install Medley