Production-grade multi-agent orchestration framework for Claude Code
aistack is a lightweight, production-ready multi-agent orchestration framework designed for Claude Code integration. It provides a complete system for managing specialized AI agents, persistent memory with semantic search, drift detection, consensus checkpoints, resource monitoring, and seamless MCP (Model Context Protocol) integration.
| Document | Description |
|---|---|
| Architecture | System architecture with C4 diagrams |
| High-Level Design | System overview and container views |
| Low-Level Design | Module breakdowns and component details |
| API Reference | Complete API documentation (46 MCP tools) |
| Data Models | Database schemas and storage details |
| Security | Security model and best practices |
| Operations | Deployment, monitoring, and troubleshooting |
| Onboarding | Getting started guide for developers |
| ADRs | Architecture Decision Records (6 ADRs) |
npm install @blackms/aistacknpx @blackms/aistack initThis creates:
aistack.config.json- Configuration filedata/- SQLite database directoryplugins/- Plugin directory
claude mcp add aistack -- npx @blackms/aistack mcp start{
"version": "1.5.3",
"memory": {
"path": "./data/aistack.db",
"defaultNamespace": "default",
"vectorSearch": {
"enabled": true,
"provider": "openai",
"model": "text-embedding-3-small"
}
},
"providers": {
"default": "anthropic",
"anthropic": {
"apiKey": "${ANTHROPIC_API_KEY}"
}
},
"agents": {
"maxConcurrent": 5,
"defaultTimeout": 300
},
"driftDetection": {
"enabled": false,
"threshold": 0.95,
"behavior": "warn",
"asyncEmbedding": true
},
"resourceExhaustion": {
"enabled": false,
"warningThresholdPercent": 0.7,
"autoTerminate": false
},
"consensus": {
"enabled": false,
"requireForRiskLevels": ["high", "medium"],
"reviewerStrategy": "adversarial"
},
"github": {
"enabled": true,
"useGhCli": true
}
}- 11 Specialized Agents: Coder, Researcher, Tester, Reviewer, Adversarial, Architect, Coordinator, Analyst, DevOps, Documentation, Security Auditor
- Persistent Memory: SQLite with FTS5 full-text search and optional vector embeddings
- 46 MCP Tools: Complete Claude Code integration across 8 categories
- 6 LLM Providers: Anthropic, OpenAI, Ollama, Claude Code CLI, Gemini CLI, Codex
- Agent Identity v1: Persistent agent identities with lifecycle management (created → active → dormant → retired)
- Semantic Drift Detection: Embedding-based similarity checking to prevent task duplication
- Resource Exhaustion Monitoring: Per-agent tracking with configurable thresholds and auto-pause
- Consensus Checkpoints: Risk-based gating for high-stakes task creation
- Adversarial Review Loop: Automatic code improvement through iterative feedback
- Plugin System: Extend agents, tools, hooks, and providers
- Task Coordination: Priority queue with hierarchical orchestration
- Workflow Engine: Multi-phase workflows with adversarial validation
- Web Dashboard: Real-time monitoring with React 18 + Material-UI
| Category | Count | Description |
|---|---|---|
| Agent | 6 | Spawn, list, stop, status, types, update |
| Identity | 8 | Create, get, list, update, activate, deactivate, retire, audit |
| Memory | 5 | Store, search, get, list, delete |
| Task | 8 | Create, assign, complete, list, get, drift check, relationships, metrics |
| Consensus | 5 | Check, list pending, get, approve, reject |
| Session | 4 | Start, end, status, active |
| System | 3 | Status, health, config |
| GitHub | 7 | Issues (create/list/get), PRs (create/list/get), repo info |
Total: 46 MCP Tools
- Node.js 20.0.0 or higher
- SQLite 3 (bundled via better-sqlite3)
- Optional: GitHub CLI (
gh) for GitHub operations - Optional: Ollama for local embeddings
Current version: 1.5.3
MIT License - see repository for details.