picoclaw.io

What It Is

PicoClaw is an ultra-lightweight personal AI assistant written in Go. It is not positioned as a proprietary large language model, but rather as a self-hosted AI assistant framework that can run on Raspberry Pi, RISC-V, Linux, Android, Docker, or cloud servers. Delivered as a single binary, it claims memory usage below 10MB, startup in under 1 second, and operation with as little as 64MB RAM, making it suitable for low-cost hardware and edge scenarios.

Core Capabilities

On the model side, PicoClaw supports OpenAI, Claude, DeepSeek, Gemini, Mistral, Doubao, Tongyi Qianwen, Kimi, Ollama, and any OpenAI-compatible API, allowing users to switch providers through configuration. On the channel side, it covers 16+ platforms, including Telegram, Discord, Slack, DingTalk, Feishu, WeCom, LINE, and QQ. It also provides web search, long-term memory, scheduled tasks, cron automation, and a Gateway mode, enabling it to serve as an AI backend via the MCP protocol and REST API.

Pricing and Privacy

PicoClaw is free and open source, and the main text says it uses an MIT-compatible license. The actual cost comes from the LLM APIs selected by the user or the resources required to run local models; no built-in free quota is provided. On privacy, it emphasizes self-hosting: configuration and chat history remain local, with no need to rely on PicoClaw cloud services. However, if cloud-based models such as OpenAI or Claude are used, input content will still be sent to the respective providers, so compliance needs to be assessed separately.

Pros, Cons, and Who It’s For

Its strengths are extremely low resource usage, cross-platform support, simple deployment, broad model and chat-channel coverage, and relatively strong support for the Chinese ecosystem. The downsides are that users need to configure API keys, runtime environments, and channel bots themselves, making it less effortless for non-technical users than SaaS products like ChatGPT. Output quality mainly depends on the connected model, and the source text does not disclose enterprise-grade permissions, auditing, or SLA features. It is best suited for developers, hardware tinkerers, self-hosting enthusiasts, and users who want to integrate AI into private devices or team chat tools.

Access from China

The source text does not specify network accessibility from mainland China, so it should be considered unknown. It supports Chinese-language pages, Feishu forms, WeChat groups, WeChat Official Accounts, Xiaohongshu, as well as ecosystems such as DingTalk, Feishu, WeCom, QQ, Doubao, Tongyi Qianwen, Kimi, and DeepSeek. Payment information is not disclosed. If external models are unavailable, users can consider connecting domestic OpenAI-compatible APIs or local Ollama models as alternatives.