🐝 Swarm AI

Cross-agent user profile synchronization. Teach one agent about yourself — every agent knows.

The Problem

You use 5 AI agents. Each one asks the same questions. You repeat yourself endlessly. Swarm fixes this with a shared profile layer that any agent can read and write.

Live Demo

Try it without installing: hive.peonai.net

Test account: peon / 123456

⚠️ Privacy Warning: This is a shared public instance. Do NOT connect your real AI agents or enter personal information. Use a disposable agent or VM for testing.

Quick Start

# Install & setup (interactive)
npx @peonai/swarm

# Uninstall
npx @peonai/swarm uninstall

Connect an Agent

Option A: MCP (Recommended for Claude Code, Cursor, Windsurf)

Add to your project's .mcp.json:

{
  "mcpServers": {
    "swarm-ai": {
      "command": "npx",
      "args": ["@peonai/swarm", "mcp"],
      "env": {
        "SWARM_API_URL": "https://your-swarm-instance.com",
        "SWARM_API_KEY": "your-agent-key"
      }
    }
  }
}

Your agent gets 6 tools: read_profile, update_profile, observe, search_memory, write_memory, read_persona. See docs/mcp.md for details.

Option B: llms.txt (For agents that support URL-based onboarding)

1. Open dashboard → create admin account
2. Copy the onboarding prompt from Dashboard or Agents page
3. Send it to any AI agent
4. Done. The agent reads llms.txt, learns the API, and starts syncing.

Features

• MCP Server — Native integration with Claude Code, Cursor, Windsurf via Model Context Protocol
• Multi-user — Setup wizard, JWT login, user management
• User profiles — Layered (identity, preferences, work, context), with confidence scores
• Shared memory — Cross-agent memories with FTS5 full-text search
• Semantic search — Optional embedding API (OpenAI-compatible)
• Agent personas — Per-agent personality and instructions
• llms.txt — Dynamic API docs that agents can self-read to onboard
• Tenant isolation — All data scoped by user
• Audit log — Track every API action and profile change
• Admin controls — Disable users, reset tokens, localhost-only admin API
• Mobile-friendly — Responsive dashboard with collapsible sidebar
• SQLite / PostgreSQL — SQLite for dev, Postgres for scale
• Webhooks — Real-time notifications when profiles or memories change
• OpenAPI spec — Machine-readable API docs at /api/openapi
• Python SDK — pip install swarm-ai-client
• Observe API — Auto-extract profile data from natural language

API

GET  /api/v1/profile          # Read user profile
PATCH /api/v1/profile         # Write profile entries
POST /api/v1/profile/observe  # Auto-extract from text
GET  /api/v1/memory           # Search memories
POST /api/v1/memory           # Write memory
GET  /llms.txt?key=<token>    # Self-service API docs for agents
POST /api/v1/webhooks         # Register webhook
GET  /api/v1/webhooks         # List webhooks
GET  /api/openapi             # OpenAPI 3.0 spec
GET  /api/health              # Health check

Auth: Authorization: Bearer <api_key> (agent key or user token)

Architecture

Agent A ──┐                    ┌── Profile (layered)
Agent B ──┤── Swarm API ──────┤── Memory (FTS5)
Agent C ──┘   (REST + JWT)     └── Audit Log

• npm: @peonai/[email protected]
• Stack: Next.js standalone, SQLite/PostgreSQL
• Deploy: systemd service via interactive CLI

License

MIT