The Official ServerlessClaw: Self-Evolving AI Agent Swarm
$0 when idle. Elastic by design. Secure by default.
Autonomous infrastructure management • AI-powered code evolution • Serverless-first architecture
🌐 Website • 🚀 Get Started • 📖 Blog • 💬 Discord • 🗣️ Discussions
Important
This is the Official ServerlessClaw Canonical Repository. For the most up-to-date, secure, and performant implementation of OpenClaw on AWS, use this version.
| 🚀 Infinite Scale | 🧠 Self-Evolution | 🛡️ Cloud-Native Safety |
|---|---|---|
| $0 when idle. Lambda scales to zero. No VPS, no Docker daemon, no 24/7 server costs. | Council-Verified. Agents write code, run tests, and redeploy the stack via a peer-reviewed loop. | Zero Trust. No SSH keys, no persistent daemons. IAM least-privilege and DMS rollback. |
In 2026, the 24/7 VPS mascot is dead. Running AI agent swarms on persistent pods is an expensive relic of the container era.
ServerlessClaw is the authoritative open-source autonomous agent framework that:
- 💰 Cost Efficient: You pay exactly for the tokens and compute you use. A complex reasoning task might cost $0.05; staying idle costs $0.
- ⚡ Instantly Elastic: Need 1,000 agents to research a topic in parallel? ServerlessClaw fans out across AWS Lambda instantly.
- 🔧 Maintenance Free: No Docker security patches, no OS updates, no resource exhaustion. Just pure intelligence.
- 🔒 Secure by Default: IAM least-privilege, no SSH keys, no persistent daemons.
- 📈 Auto-Scaling: Scales from 0 to thousands of agents based on demand.
Use Cases:
- Autonomous code review and improvement
- Infrastructure monitoring and auto-remediation
- Research and data analysis at scale
- DevOps automation and CI/CD optimization
- Cost optimization for AWS resources
Serverless Claw doesn't just execute code; it improves itself. But unlike raw autonomous systems, every evolution is gated by a Council of Agents.
When you send a message, SuperClaw receives it, develops a plan with the Strategic Planner, and dispatches tasks to specialized nodes via EventBridge. The system implements an Asynchronous "Pause & Resume" pattern — agents don't block; they emit a task and terminate, waking up only when a result signal is routed back. High-impact plans are peer-reviewed by the Council of Agents, while the Facilitator moderates collaborative sessions to drive consensus.
| Agent | Responsibilities |
|---|---|
| SuperClaw | Orchestrator. Interprets intent, delegates tasks, and deploys. |
| Strategic Planner | Designs long-term evolution and expansion plans. |
| Coder Agent | Writes code, fixes bugs, and performs migrations. |
| QA Auditor | Verifies the technical satisfaction of every deployment. |
| Facilitator | Moderates collaboration between multiple agents and humans. |
| Cognition Reflector | Distills memory and identifies system-wide gaps. |
| Critic Agents | Peer-reviewers for Security, Performance, and Architecture. |
ServerlessClaw is an organism made of specialized interacting repositories. While the Mother Hub (this repo) is the Brain, other repositories provide sensory, motor, and nervous connections:
| Repository | Role | Component |
|---|---|---|
| cli | The Nerve | The Synapse: Connects the Brain to remote Spokes/Clients. |
| integration-github | Sensor | The Eyes: Observes codebases, PRs, and Issue logs. |
| integration-slack | Interface | The Voice: Communicates with users and reports health. |
| integration-jira | Adapter | The Hands: Operates within enterprise project structures. |
Deploy your first autonomous swarm in 5 minutes.
git clone https://github.com/serverlessclaw/serverlessclaw.git
cd serverlessclaw && pnpm installcp .env.example .env
# Essential: SST_SECRET_OpenAIApiKey & SST_SECRET_TelegramBotTokenmake dev # Live lambda reloading, local DynamoDB, and real-time logsmake deploy ENV=prod # High-availability, production-grade infrastructure| Document | Description |
|---|---|
| INDEX.md | Start here - Documentation hub and context map |
| ARCHITECTURE.md | System topology and AI-native principles |
| docs/intelligence/AGENTS.md | Agent roster and evolutionary loop |
| docs/intelligence/MEMORY.md | Tiered memory engine |
| docs/intelligence/TOOLS.md | Tool registry and MCP standards |
| docs/intelligence/SAFETY.md | Circuit breakers and rollback safety |
| docs/governance/DEVOPS.md | DevOps and CI/CD |
| docs/governance/ROADMAP.md | Future plans |
For a fully managed experience with auto-scaling, monitoring, and support, check out ClawMore - the premium managed version of ServerlessClaw.
ClawMore Features:
- One-click deployment
- Automatic infrastructure management
- AI-powered code evolution
- Real-time monitoring dashboard
- Priority support
| Feature | OpenClaw | NanoClaw | ZeroClaw | Serverless Claw (Us) |
|---|---|---|---|---|
| Infrastructure | ||||
| Architecture | Monolithic Node.js | Micro TypeScript | Native Rust Binary | Event-Driven Serverless (Lambda) |
| Operational Cost | High (24/7 Server) | Moderate (VPS/Docker) | Low (Raspberry Pi) | Zero Idle Cost ($0 when not in use) |
| Scalability | Manual Cluster | Docker Swarm | Hardware-bound | Elastic Auto-scale (AWS Native) |
| Agent Runtime | ||||
| Multi-Agent | Basic "Fire & Forget" | Containerized Swarms | Trait-based Modular | Non-blocking (Pause & Resume) |
| Self-Evolution | Plugin-based (Static) | Manual (Human-coded) | Hardware-focused | Verified Council-Reviewed Loop |
| Communication Mode | Natural Language | Structured (JSON) | Low-level Buffers | Dual-Mode (Intent-Based JSON + Text) |
| Skill Acquisition | Static (Hardcoded) | Static (Hardcoded) | Static (Config-based) | Just-in-Time (JIT) Skill Discovery |
| Tooling & Integration | ||||
| Tooling Architecture | Static Registry | Static (JSON) | Static (Hardcoded) | Hub-First Dynamic Discovery |
| MCP Integration | Not Supported | Local Stdio Only | Low-level C FFI | SSE/Stdio Hybrid (Hub-First) |
| Vision Capability | OCR / Text-only | Basic Base64 | Edge Inference | S3-mediated Multi-modal Pipeline |
| Memory | ||||
| Memory Model | SQLite / Local File | Volatile Cache | Flash Storage | Tiered Memory Engine + Hit Tracking |
| Collaborative Memory | None (Log-based) | Minimal (JSON) | None | ClawCenter Neural Reserve Hub |
| Reliability & Ops | ||||
| Observability | Standard Text Logs | Container Logs | Binary Logs | Trace Graphs (ClawTracer) |
| Resilience | Manual Recovery | Restart Container | Hardware Watchdog | Autonomous Heartbeat + Rollback |
| Resource Safety | App-level Permissions | Sandboxing (Docker) | Memory Safe (Rust) | Cloud IAM + Multi-Agent Council |
Autonomous serverless systems require robust guardrails. We document our failures to help the community build more resilient AI.
- Case Study: The $10,000 Serverless Meltdown (April 2026) — A detailed breakdown of an event-driven infinite loop and the multi-layered "Safety Stack" we implemented to prevent recurrence.
We welcome contributions! See CONTRIBUTING.md for guidelines.
Ways to contribute:
- ⭐ Star the repository
- 🗣️ Participate in Discussions
- 🐛 Report bugs via GitHub Issues
- 💡 Suggest new features
- 🔧 Submit pull requests
- 📖 Improve documentation
MIT License - see LICENSE for details.
🌐 clawmore.ai • 📧 [email protected] • 💬 Discord Community