PicoClaw
Officially listedAn ultra-lightweight AI agent framework that runs on $10 hardware with under 10MB of memory.
PicoClaw
PicoClaw is an ultra-lightweight open-source AI agent framework from Sipeed, built from scratch in Go and capable of running on RISC-V development boards costing as little as $10. With memory usage under 10MB and startup in under 1 second, it is built for edge devices and resource-constrained environments.
Core Capabilities
- Extreme lightness: Memory footprint under 10MB—99% less than OpenClaw—with cold start within 1 second
- All-architecture coverage: A single binary supports RISC-V, ARM, x86, MIPS, and nearly every other CPU architecture
- Native MCP integration: Supports the Model Context Protocol for flexible extension of the agent toolchain
- Multi-platform chat: Connects to Telegram, Discord, WhatsApp, QQ, DingTalk, and 10+ other chat platforms
- Multi-model support: Compatible with mainstream LLM providers including OpenRouter, Anthropic, OpenAI, and Ollama
- Offline-first: Task queues persist to SQLite, so local tasks keep running even when disconnected
Use Cases Embedded/IoT developers, edge computing engineers, Raspberry Pi hobbyists, and self-hosting enthusiasts. If you want to run an AI assistant at the lowest possible cost, this is the answer.
Unique Advantages Software made by hardware vendor Sipeed, naturally fitted to low-cost edge devices. Where OpenClaw needs 1GB of memory and $400 hardware, PicoClaw runs on 10MB and $10.
Editor's Review It gathered 25K+ GitHub Stars within two months of release—community momentum is fierce. In the lightweight AI agent lane it currently has no rival. Still, it is early-stage (v0.2.3), the team does not yet recommend production use, and it is wise to wait for v1.0 before serious deployment. Recommendation: 4/5.
Pricing
### 💰 定价模式：免费开源 **起步价**：免费（MIT 许可证） #### 主要方案 - **软件**：$0 - 完全免费开源，无功能限制 - **最低硬件**：$9.9 - LicheeRV-Nano RISC-V 开发板 - **推荐硬件**：Raspberry Pi Zero 2 W 或 Pi 5 #### 试用/其他信息 也可在任何 Linux/macOS/Windows 电脑上运行。LLM API 费用需自行承担，支持 Ollama 本地模型实现完全免费离线运行。
FAQ
Is PicoClaw free?
It is completely free and open source under the MIT license, with no feature restrictions. But running PicoClaw means covering your own LLM API costs—or using local Ollama models for zero-cost operation.
What are PicoClaw's main features?
Core features include an ultra-lightweight AI agent runtime (<10MB memory), MCP protocol support, multi-chat-platform integration, multi-LLM-provider compatibility, scheduled tasks, multi-agent coordination, vision pipelines, and an offline task queue.
What hardware does PicoClaw need to run?
The minimum is a $9.9 LicheeRV-Nano RISC-V development board; Raspberry Pi Zero 2 W or Pi 5 are recommended. It also runs on any Linux, macOS, or Windows computer—no special hardware required.
How does PicoClaw differ from OpenClaw?
PicoClaw uses under 10MB of memory (OpenClaw needs 1GB+), starts 400x faster, and runs on $10 hardware. But OpenClaw is more feature-rich, supports advanced plugins like browser automation, and has a more mature community.
Is PicoClaw ready for production environments?
Not recommended yet. PicoClaw is still in early development (latest version v0.2.3), and the team explicitly states there are unresolved security issues. Wait for the v1.0 stable release before production use.
Which chat platforms does PicoClaw support?
It supports 10+ chat platforms including Telegram, Discord, WhatsApp, Matrix, IRC, QQ, DingTalk, LINE, and WeCom. Telegram also supports voice messages.
Can PicoClaw run offline?
Yes. With local Ollama models and the PicoLM inference engine, PicoClaw supports fully offline operation. Task queues persist to SQLite, so local tasks still execute when the network is down.