Autonomous AI agents that run 24/7, learn from every task, and coordinate with each other.
Security-hardened fork of OpenClaw · Open source · MIT
Website · Documentation · Quick Start · Premium · GitHub
Klawty is an operating system for AI agents. It's a security-hardened fork of OpenClaw (302K+ stars, MIT) with custom security modules built by dcode technologies.
The free version gives you 1 agent with the full OpenClaw platform (40+ CLI commands, 20+ channels, 52 skills, plugin SDK) plus Klawty's security layer (policy engine, exec sandbox, PII detection, credential monitoring, runtime integrity).
For multi-agent teams, smart routing, industry skills, and managed hosting, see the premium version.
curl -fsSL https://klawty.ai/install.sh | bash# Option 1: One-command install
curl -fsSL https://klawty.ai/install.sh | bash
# Option 2: From source
git clone https://github.com/dcode-tec/Klawty.git
cd Klawty && pnpm install && pnpm build
klawty onboard # interactive setup
klawty run # start your agent
# Option 3: Docker
git clone https://github.com/dcode-tec/Klawty.git
cd Klawty && docker compose up -dRequirements: Node.js 22+ · OpenRouter API key (recommended)
| Layer | Source | What it provides |
|---|---|---|
| Platform | OpenClaw (MIT) | CLI, 20+ channels, plugin SDK, gateway, memory, native apps, 52 skills |
| Security | dcode technologies | Policy engine, Docker exec sandbox, PII detection, credential monitoring, integrity checks |
| Agent Runtime | Klawty (premium) | 5-tier LLM routing, 4-tier memory, proposals, dedup, self-improvement |
| Feature | Details |
|---|---|
| 1 agent (Atlas) | General-purpose orchestrator with read-only tools |
| CLI (40+ commands) | klawty run, status, stop, tui, logs, onboard, doctor, ... |
| 20+ channels | Discord, Slack, Telegram, WhatsApp, Signal, Matrix, IRC, Teams, ... |
| 52 community skills | Web research, coding, data analysis, comms, devops, ... |
| Plugin SDK | Build custom skills and channel integrations |
| Security policy | klawty-policy.yaml — deny-by-default (network, filesystem, exec) |
| Docker exec sandbox | Shell commands in isolated containers (no network, read-only) |
| PII detection | Email, phone, credit card, IBAN auto-detect + local routing |
| Credential monitor | API key validation every 6 hours, balance alerts |
| Runtime integrity | SHA-256 manifest verified on every boot |
| TUI dashboard | Full-screen terminal: agents, tasks, costs, health |
| Docker deployment | docker-compose.yml included |
| File-based memory | MEMORY.md (50 lines) — persistent across restarts |
Available at ai-agent-builder.ai:
- Up to 8 coordinated agents with inter-agent messaging
- 5-tier LLM routing — 10x cost savings via smart model selection
- 4-tier memory — file + session + JSONL + Qdrant vector search
- 27 domain skills — SEO, copywriting, sales, finance, client ops, ...
- Self-improvement — reflection engine + skill gap detection
- Proposal lifecycle — approval + 15-minute rollback
- 4-layer deduplication — task, channel, proposal, discovery
- Management dashboard — web portal with KPIs, kanban, costs
- 6 industry solutions — restaurants, real estate, construction, resellers, accounting, law firms
- Managed hosting — all-inclusive from 79€/month (hosting + AI credits + support)
klawty run # start agents
klawty status # health + tasks at a glance
klawty stop # graceful shutdown
klawty tui # full-screen terminal dashboard
klawty logs [agent] # tail agent logs
klawty onboard # interactive setup wizard
klawty doctor # diagnose issues
klawty agent --message # run one agent turn
klawty channels list # connected channels
klawty memory search # search agent knowledge
klawty security audit # scan for vulnerabilities
klawty backup create # snapshot state
klawty update # check for updates
klawty plugins list # loaded plugins (36+)
klawty skills list # available skills (52+)
klawty models list # configured models
klawty --version # Klawty 1.0.0All security modules live in src/security/ — real code, not templates.
| Module | Lines | What it does |
|---|---|---|
policy-enforcer.js |
493 | Reads klawty-policy.yaml, enforces network/filesystem/exec rules |
exec-sandbox.js |
229 | Runs shell commands in Docker containers (no network, read-only root) |
privacy-router.js |
211 | Detects PII (email, phone, IBAN) → local model routing or redaction |
credential-monitor.js |
448 | Validates API keys every 6h, alerts on expiry or low balance |
integrity-check.js |
188 | SHA-256 manifest of all modules, verified on every boot |
auto-update.js |
470 | Version check + download + rollback on failure |
tui-dashboard.js |
624 | Full-screen ANSI dashboard: agents, tasks, costs, health |
klawty-boot.js |
95 | Boot hook — initializes all security modules on gateway start |
Total: 2,758 lines of custom security code.
workspace/
├── klawty.json # main config (1 model, basic settings)
├── klawty-policy.yaml # security policy (deny-by-default)
├── SOUL.md # agent personality
├── IDENTITY.md # agent identity (Atlas 🦞)
├── AGENTS.md # roster
├── TOOLS.md # 5 read-only tools
├── MEMORY.md # persistent knowledge
├── HEARTBEAT.md # check schedule
├── USER.md # operator profile
├── agents/main/AGENT.md # Atlas config
├── skills/web-research/ # 1 demo skill
└── .env.example # API key template
| Stable | Beta |
|---|---|
| Discord · Slack · Telegram · WhatsApp · Signal · Matrix · MS Teams · IRC · Line · Nostr · Terminal · Web | iMessage (macOS) · Google Chat · Mattermost · Twitch |
| Resource | Minimum | Recommended |
|---|---|---|
| Node.js | 22+ | Latest LTS |
| RAM | 2 GB | 8 GB |
| Docker | Optional | Recommended (for exec sandbox) |
| OS | macOS, Linux, WSL2 | macOS or Ubuntu 22.04+ |
- Bug fixes — open an issue, then a PR
- New skills — add
SKILL.mdtoskills/ - Security reports — email security@klawty.ai
git clone https://github.com/dcode-tec/Klawty.git
cd Klawty && pnpm install && pnpm build && pnpm test- OpenClaw — 302K+ stars, MIT license. The fastest-growing open-source AI agent framework.
- Qdrant — Open-source vector database (premium feature).
- OpenRouter — Unified API for 200+ LLM models.
Built by dcode technologies · Luxembourg 🇱🇺
Security-hardened fork of OpenClaw. MIT license preserved.