The Claw Family at a Glance
The OpenClaw ecosystem offers AI agents at every scale — from a full-featured platform running on powerful servers to a tiny agent fitting inside a $2 microcontroller. But which one should you use for your research?
Full Comparison Table
| Feature | OpenClaw | NanoBot | ZeroClaw | PicoClaw | NanoClaw | MicroClaw |
|---|---|---|---|---|---|---|
| Language | TypeScript | Python | Rust | Go | TypeScript | Rust |
| Stars | 323K+ | 34K+ | 27K+ | 25K+ | 24K+ | 576 |
| Min RAM | ~512MB | ~50MB | ~30MB | <10MB | ~100MB | <1MB |
| Boot time | ~5s | ~2s | <1s | <1s | ~3s | <0.1s |
| Install | npm install -g openclaw | pip install nanobot-ai | Single binary | Single binary | Docker container | Firmware flash |
| Messaging | 20+ platforms | 9 platforms | 20+ platforms | All major | 6 platforms | N/A |
| MCP Support | Yes | Yes | Yes | Yes | Yes | No |
| Skills | ClawHub (6300+) | Compatible | Compatible | Compatible | Compatible | Limited |
| Memory | Yes | Yes | Yes | Limited | Yes | No |
| Multi-model | Yes | Yes (7+ providers) | Yes | Yes | Yes (Anthropic SDK) | Single model |
| Best for | Full-featured lab | Python researchers | Infrastructure | Edge/IoT | Secure containers | Embedded sensors |
When to Choose Each
OpenClaw — "I want everything"
Choose if: You need the full platform with all 20+ messaging integrations, ClawHub skills marketplace, Canvas rendering, and voice support. Your research lab has a server or cloud instance to run it on.
Typical user: Research lab with a shared server, wanting to deploy a team-wide AI assistant accessible from any messaging app.
npm install -g openclaw@latest
openclaw onboardNanoBot — "I'm a Python researcher"
Choose if: You work primarily in Python, want quick setup, and need multi-model support (OpenAI, Anthropic, Gemini, Ollama). You don't need 20 messaging platforms — 9 is plenty.
Typical user: Bioinformatics PhD student who wants an AI assistant in their existing Python environment, connected to Slack or Telegram.
pip install nanobot-ai
nanobot onboardKey advantage: 99% less code than OpenClaw while keeping all core agent capabilities. If you're already in a Python conda/venv environment, this is the path of least friction.
ZeroClaw — "I need performance and autonomy"
Choose if: You're deploying AI agent infrastructure at scale, need Rust-level performance, and want fully autonomous operation with minimal supervision.
Typical user: Research IT team setting up self-hosted AI infrastructure for an entire department.
PicoClaw — "I'm deploying on edge hardware"
Choose if: Your research involves IoT, field sensors, or embedded devices. You need an AI agent that runs on $10 hardware with <10MB RAM.
Typical user: Environmental researcher deploying data-collection agents on Raspberry Pi or RISC-V boards in the field.
Supported architectures: x86_64, ARM64, MIPS, RISC-V, LoongArch
NanoClaw — "Security is my priority"
Choose if: You handle sensitive data (patient records, proprietary compounds) and need container-isolated execution. Built on Anthropic's Agents SDK.
Typical user: Clinical researcher or pharmaceutical company needing HIPAA-aware AI assistance.
docker pull nanoclaw/nanoclaw
docker run nanoclaw/nanoclawMicroClaw — "I'm going embedded"
Choose if: You're building AI into microcontrollers (ESP32, Arduino) for lab automation or sensor networks. Runs in <1MB RAM.
Typical user: Lab automation engineer adding AI decision-making to instrument firmware.
Decision Flowchart
Need 20+ messaging platforms?
→ Yes: OpenClaw
→ No: Continue
Python ecosystem?
→ Yes: NanoBot
→ No: Continue
Need container isolation / security?
→ Yes: NanoClaw
→ No: Continue
Deploying on edge/IoT hardware?
→ Yes: PicoClaw (<10MB) or MicroClaw (<1MB)
→ No: Continue
Need Rust performance at scale?
→ Yes: ZeroClaw
→ No: Start with NanoBot (simplest)Compatibility
All frameworks in the Claw family support the same SKILL.md skill format and MCP (Model Context Protocol), meaning:
- Skills from ClawHub, K-Dense, or LabClaw work across all platforms (except MicroClaw)
- MCP servers can be shared between agents
- You can start with NanoBot and migrate to OpenClaw later without rewriting skills