A complete AI-powered development environment featuring 12 specialized agents, 63+ skills, and workflow orchestration with intelligent routing, parallel execution, and MCP integration.
- Node.js 18+ with npm
- lmstudio (optional, for local models)
- OpenCode CLI installed globally
# Install dependencies
npm install
# Start OpenCode (recommended method)
npm start
# Or on Windows:
opencode.bat- Switch to an agent:
/agent backend-laravel - Get agent recommendation: Ask "Which agent should handle Laravel auth?"
- Run commands:
/build,/test,/lint - Use workflows: Ask lead-strategist to "Add user authentication feature"
All documentation is centralized in the docs/ folder:
| Document | Description |
|---|---|
| User-Guide.md | Basic usage, features, quick start |
| Workflows-Guide.md | Multi-agent workflow automation |
| Skills-Guide.md | 63+ specialized capabilities with MCP integration |
| Plugins-Guide.md | 10+ plugins including agent-router |
| Brain-Plugin-Docs.md | RAG system with local embeddings and Rust sidecar |
| Agents-Guide.md | Complete agent reference (12 agents) |
| Agent-Loop-Guide.md | Iterative execution and retry patterns |
| Prompting-Guide.md | Effective prompt techniques |
| Prompting-and-Context-Engineering.md | Context engineering deep dive |
OpenCode uses a Swarm Architecture where specialized agents collaborate to solve problems:
lead-strategist (Orchestration)
├── Backend (api, laravel, tauri)
├── Frontend (ui-ux)
├── QA/DevOps (guardian, engineering)
├── Core (core-factory)
└── Docs (docs-curator)
| Category | Agents | Specialization |
|---|---|---|
| Core | core-factory | Core implementation |
| Lead | lead-strategist, lead-architect | Orchestration, Architecture |
| Backend | backend-api, backend-laravel, backend-tauri | API, Laravel, Tauri |
| Frontend | frontend-ui-ux | UI/UX Engineering |
| QA | qa-guardian | Review, Testing, Security, Debug |
| DevOps | devops-engineer | Operations, MCP |
| Docs | docs-curator | Writing, Governance, Evolution |
Automatic codebase context injection using local embeddings:
- Local Embeddings: nomic-embed model via LM Studio (no cloud)
- Rust Sidecar: High-performance vector search with HNSW
- Tree-sitter Chunking: AST-aware code splitting for semantic chunks
- Intent Classification: 7 intent types with adaptive retrieval strategies
How it works:
- User types query → Intent classified (debug/learn/feature/etc.)
- Relevant code chunks retrieved via vector search
- Context prepended to LLM prompt for grounded responses
Status: TS plugin loads successfully; Rust sidecar requires Linux/Mac for building.
Complex multi-step tasks with advanced features:
phases:
- name: Strategy & Analysis
use_agent_router: true
mcp_tools:
context7: [fetch_library_docs]
memory: [create_entities]
parallel_groups:
- [research_auth_methods, analyze_existing_patterns]Key Features:
- Parallel Execution:
parallel_groupsfor independent tasks - Retry Policies: Configure
max_attemptsand backoff strategies - MCP Integration: Declarative
mcp_toolsper phase - Performance Tracking: Metrics in sqlite MCP
- Security Scanning: Automated vulnerability scans
- Ambient LSP Feedback: Automatic error detection after edits
8+ Model Context Protocol servers for enhanced capabilities:
| Server | Purpose | Timeout |
|---|---|---|
context7 |
Up-to-date documentation | 30s |
filesystem |
File system operations | 15s |
memory |
Persistent knowledge graph | 10s |
git |
Git repository operations | 20s |
fetch |
Web content fetching | 15s |
sqlite |
SQLite database operations | 10s |
sequential-thinking |
Step-by-step reasoning | 30s |
language-server |
LSP integration | 20s |
Skills provide domain-specific capabilities:
- Development:
laravel-feature-scaffold,fullstack-dev,stack-context - Content:
blog-writer,docs-governance-audit,deep-research - Analysis:
security-review,testing-strategy,database-design - Automation:
workflow-manager,self-improver,project-orchestration - Media:
pdf,ppt,xlsx,docx,image-generation,podcast-generate - Research:
market-research-reports,aminer-open-academic,ai-news-collectors
opencode/
├── opencode.json # Main configuration
├── package.json # Dependencies
├── README.md # This file
├── CHANGELOG.md # Release history
├── CONTRIBUTING.md # Contribution guide
├── SECURITY.md # Security policy
├── CODE_OF_CONDUCT.md # Code of conduct
├── LICENSE # MIT License
├── .gitignore # Git exclusions
├── .env.example # Environment template
├── tsconfig.json # TypeScript config
├── vitest.config.ts # Test configuration
│
├── agents/ # 12 specialized agent definitions
├── skills/ # 63+ specialized capabilities
├── plugins/ # 10+ TypeScript plugins
├── rules/ # Code style guidelines
├── workflows/ # YAML workflow definitions
├── docs/ # Centralized documentation
├── tools/ # Utility scripts
├── scripts/ # Build & deployment scripts
├── .github/ # CI/CD workflows & templates
└── docs/adr/ # Architecture Decision Records
- Model:
opencode-go/kimi-k2.6(default) - Providers: opencode, lmstudio, cerebras, opencode-go
- Agents: 12 custom agents with specialized tools
- MCP Servers: 8+ servers with timeouts
- Plugins: 10+ plugins (agent-router, model-router, etc.)
- Permissions: Granular tool permissions, sensitive file protection
- Primary:
npm start(recommended) - Cross-platform:
opencode.bat(Windows),opencode.sh(Unix/macOS) - JavaScript:
node opencode-launch.js(any platform)
See CONTRIBUTING.md for detailed guidelines.
Quick reference:
- Add a skill: Create
skills/my-skill/SKILL.md, register inskills/index.json - Add a plugin: Create
plugins/my-plugin.ts, register inopencode.json - Add a workflow: Create
workflows/my-workflow.yamlwith v2.0.0 schema - Run tests:
npm test(Vitest suite)
- Root directory with ~15 essential files
- Build artifacts and dependencies in
.gitignore - Centralized documentation in
docs/ - Consistent structure with clear separation of concerns
MIT License - See LICENSE file for details.
- Documentation: Check
docs/folder - Issues: Report bugs via GitHub Issues
- Discussions: Join community discussions
Last Updated: 2026-05-08 Version: 2.0.0 (Workflow Orchestration)