/mission for Claude Code

Product Introduction

1. Definition: Medley is a free, multi-agent orchestration plugin for Claude Code, a terminal-based AI development environment. Technically, it is a plugin that decomposes complex user prompts into structured, executable task graphs and coordinates multiple AI agents (using various LLMs) to execute those tasks in sequence or parallel.
2. Core Value Proposition: It exists to overcome the limitations of single-session, single-model AI assistance by enabling coordinated multi-agent execution for complex, multi-step projects directly within the Claude Code terminal. Its primary value is turning a high-level outcome into a managed, visible workflow.

Main Features

1. Intelligent Prompt Decomposition (/mission command): When a user types /mission followed by a goal (e.g., "Find a GTM channel"), Medley's core engine analyzes the prompt and breaks it down into a hierarchical task tree. This tree defines dependencies, assigns agents, and sets handoffs. It outputs this plan both in the terminal and as a unique, shareable URL for a web-based mission board.
2. Multi-Model, Multi-Agent Orchestration: Medley does not rely on a single AI model. Its orchestration layer can assign different tasks to different AI agents powered by various models (e.g., Claude for research, GPT for list building, Gemini for drafting). This allows for parallel execution and leveraging specialized model strengths, a key tenet of multi-agent AI systems.
3. BYOK (Bring Your Own Key) Integration via OpenRouter: A critical technical feature is its model-agnostic support via OpenRouter integration. Users can bring their own API keys for a wide range of LLMs beyond Claude, including Kimi, GLM, DeepSeek, Qwen, GPT, Gemini, and local models via Ollama. This provides flexibility and cost control.
4. Extensive Tool and Platform Connectivity: The plugin offers pre-built integrations with a vast ecosystem of developer and productivity tools. This includes GitHub, GitLab, VS Code, Vercel, AWS, Supabase, Docker, PostgreSQL, as well as apps like Slack, Notion, Gmail, Google Drive, Linear, Jira, and Figma. This allows agents to perform real-world actions within these platforms.

Problems Solved

1. Pain Point: The context window and single-session limitation of standard AI coding assistants. Claude Code alone operates within one session; complex projects that require research, planning, execution, and review across different domains exceed this scope.
2. Target Audience: Senior Developers, Engineering Managers, Startup Founders, and Growth Hackers who use Claude Code for serious project work. Specifically, developers managing complex full-stack features, founders orchestrating go-to-market experiments, and technical operators automating multi-step workflows between code and business apps.
3. Use Cases:
   • Technical Project Orchestration: Automating the steps for a new feature: research, code implementation, testing, documentation, and creating a Linear ticket.
   • Go-to-Market (GTM) Execution: From a prompt like "launch a waitlist," decomposing into market research, landing page copy, lead list building, and email campaign drafting.
   • Multi-Step Research & Synthesis: Conducting competitive analysis by researching multiple companies, synthesizing findings into a report, and formatting it for a Notion doc or Confluence page.
   • Overnight/Batch Processing: Setting a mission to run tasks sequentially or in parallel over an extended period, effectively acting as an AI-powered asynchronous worker system.

Unique Advantages

1. Differentiation: Unlike standalone AI agents or chatbots that operate in isolation, Medley is deeply integrated into the Claude Code terminal, the existing developer workflow. Unlike manual prompt chaining, it provides automatic decomposition, visibility (via the mission board URL), and coordinated execution. It is not a separate SaaS app but a workflow enhancer for a specific, powerful environment.
2. Key Innovation: Its "mission-first" orchestration paradigm. The innovation is not just multi-agent capability, but the specific workflow of: 1) Prompt decomposition into a visual graph, 2) Intelligent agent assignment based on task type, 3) Providing a real-time status board for human-in-the-loop review, all initiated from a simple CLI command. This creates a structured, auditable, and controllable multi-agent process.

Frequently Asked Questions (FAQ)

1. How does Medley's multi-agent approach improve results over using Claude Code alone? Medley employs a coordinated multi-agent plan where different tasks are executed by specialized AI agents, often in parallel. This breaks the single-model bottleneck, allowing for division of labor, use of optimal models per task, and handling of workflows larger than one context window, leading to higher quality and more complete outcomes as demonstrated in its benchmark leads.
2. Is Medley safe to use? Can it act autonomously without my permission? Medley operates with human-in-the-loop guardrails. By default, agents run within set boundaries, and the system is designed to ask for user permission before executing sensitive actions such as sending external communications, making public commits, spending budget, or contacting people. Users can grant increased autonomy progressively.
3. What are the costs associated with using the Medley plugin? The Medley plugin itself is free and open-source. Costs are incurred only through the usage of the underlying Large Language Models (LLMs). Since Medley uses a BYOK (Bring Your Own Key) model via OpenRouter or direct API keys, users pay directly to their chosen LLM providers (e.g., OpenAI, Anthropic, Google) based on their own usage.
4. What kind of technical benchmarks or performance data supports Medley's effectiveness? Independent benchmarks show Medley's multi-agent system outperforming single-model baselines. For example, it achieved a 93.25% task pass rate on Terminal-Bench 2.1 (leading Anthropic and OpenAI), a 71.37 mean score on HealthBench Professional, and significant leads on ViBench and DrugDiscoveryBench, validating that multi-agent beats a single model alone on complex tasks.
5. Can I use Medley with locally hosted models like Llama or through Ollama? Yes, Medley's BYOK (Bring Your Own Key) architecture supports integration with locally hosted models. By configuring Medley to use an Ollama endpoint or similar local server, you can assign tasks to agents powered by models like Llama, Mistral, or other open-source LLMs running on your own hardware, keeping data and processing local.