This project provides a sophisticated configuration for OpenCode, transforming Claude into an adaptive, multi-agent system. It addresses the challenge of managing complex development tasks by enabling agents to learn from their mistakes, decompose work, and coordinate effectively. The target audience includes developers and researchers seeking to leverage AI for more intelligent and efficient software development, offering benefits like automated strategy adaptation and cost-optimized execution.

How It Works

The system employs a multi-agent orchestration framework built on joelhooks/swarmtools. It utilizes an outcome-based learning pipeline where task execution history (via CASS) informs strategy decomposition and worker execution. Successful patterns are promoted, while failures lead to anti-pattern detection and avoidance. A key innovation is the cost-optimized architecture, distinguishing between a long-lived, expensive Coordinator and disposable, focused Workers to significantly reduce operational costs. Cross-agent memory, powered by CASS, enables semantic and full-text search across all AI agent histories, preserving learnings and solutions.

Quick Start & Requirements

1. Install OpenCode: Use the official script (curl -fsSL https://opencode.ai/install | bash) or Homebrew (brew install anomalyco/tap/opencode).
2. Clone & Install Config: git clone https://github.com/joelhooks/opencode-config ~/.config/opencode followed by cd ~/.config/opencode && bun install.
3. Install CLI Tools:
   • Swarm orchestration: npm install -g opencode-swarm-plugin (requires v0.30.0+).
   • Ollama for embeddings: Install Ollama (brew install ollama or script), then ollama serve and ollama pull nomic-embed-text.
   • Cross-agent session search (optional): npm install -g cass-search (requires v0.1.35+).
4. Verify Setup: Run swarm doctor inside OpenCode.
5. Initialize Repo: Within OpenCode, navigate to your project and run /init to generate AGENTS.md.
6. Run First Swarm: Inside OpenCode, use /swarm "Your task description".

Prerequisites: OpenCode, Node.js/npm/bun, Ollama, nomic-embed-text embedding model.

Highlighted Details

• Adaptive Learning: Agents learn from execution outcomes, promoting successful strategies and flagging/inverting anti-patterns, with a 90-day confidence decay.
• Cross-Agent Memory (CASS): Provides semantic and full-text search across all AI agent histories, enabling reuse of past solutions and learnings.
• Cost-Optimized Architecture: Employs a Coordinator/Worker model to reduce AI execution costs by approximately 70%.
• Custom Tools: Integrates 12 custom MCP tools, including a multi-language Ultimate Bug Scanner (UBS), semantic memory (Hivemind), and PDF/Markdown knowledge base (pdf-brain).

Maintenance & Community

The core innovation stems from joelhooks/swarmtools. Credits are also given to nexxeln/opencode-config for specific plugins and patterns. The README does not provide direct links to community channels (like Discord/Slack) or a public roadmap.

Licensing & Compatibility

The project is licensed under the MIT license. This permissive license allows for commercial use and integration into closed-source projects without significant restrictions.

Limitations & Caveats

This repository is an OpenCode configuration, not a standalone tool; all operations must occur within the OpenCode environment. The provided CLIs (swarm, cass) function as agent backends and are not intended for direct human interaction. The setup requires several external dependencies (OpenCode, Ollama, npm packages), which may present an initial adoption hurdle.