PicoClaw review

PicoClaw is a self-hosted personal AI assistant, not a hosted API or SaaS service. Written in Go, it deploys as a single binary on hardware ranging from small boards to cloud servers. It connects to LLM providers through their APIs; PicoClaw’s stated design keeps conversation history and configuration local.

In a stack, it sits between chat interfaces or command-line use and a model provider. Gateway mode makes it an AI backend for chat-platform connections through MCP and a REST API. The homepage lists support for 16+ chat platforms, including Telegram, Discord, Slack, QQ and DingTalk. The vendor says minimum hardware is any device with 64MB RAM and an internet connection, with 512MB recommended, and lists x86_64, ARM64, ARMv7, ARMv6, RISC-V and MIPS support. It claims under 10MB of RAM use and startup in under one second. Those are vendor claims, not a substitute for measuring your own workload.

The main integration surfaces named on the homepage are Gateway mode, MCP protocol and a REST API. The assistant connects to LLM providers through APIs, including OpenAI, Anthropic, DeepSeek, Google Gemini, Mistral, Qwen, Ollama and any OpenAI-compatible API. The implementation language is Go; the homepage does not name client SDKs for the gateway, so confirm SDK support and API details in the vendor’s docs before choosing a language for your integration.

PicoClaw is distributed as a single binary, and the site links to documentation, a changelog and GitHub. The homepage also describes Android APK installation and Termux use. One Reddit user shared a setup using the binary, a configuration file, an OpenRouter API key and Telegram; another reported building from source in Termux. Treat those as individual reports, not official install instructions. Check the docs for current commands, configuration format, authentication, API behavior and supported chat-platform setup.

The software costs $0 per month: the pricing information lists a free, open-source option under an MIT license, with self-hosting and no published monthly or annual billing. The vendor says the software is free and describes the license as MIT-compatible. Seat and usage limits for the software are not published; for a self-hosted project, you provide the compute and manage capacity yourself.

The bill grows through LLM API usage. The vendor says PicoClaw connects to providers such as OpenAI and Anthropic, and its pricing information says you pay those providers directly based on consumption. Model rates, any PicoClaw-side request limits, and total usage costs are not published. The amount therefore depends on your provider, selected model and workload; check that provider’s current pricing and track usage before deploying for multiple users. There is no published PicoClaw add-on, free trial or paid plan. For production, budget for hosting and model calls separately, and ask providers about their own rate limits and overage rules.

PicoClaw is open source, and the vendor describes its license as MIT-compatible; the pricing information lists an MIT license. It runs on infrastructure you control as a single binary, rather than requiring a vendor-hosted runtime. The homepage says conversation history and configuration remain local, while LLM requests go to the provider you configure. That means model traffic is still subject to the provider’s data handling and retention terms.

The vendor does not describe a dedicated data-export feature. Because the assistant is self-hosted and the code is open source, you control its local files and can inspect the repository, but confirm the current storage format and migration steps in the docs before relying on them. Provider-specific configuration, prompts and chat-platform integrations may take work to move. Lock-in is less about a hosted subscription and more about the code and operational effort invested in your setup.

PicoClaw’s low memory and hardware claims make it worth evaluating on constrained devices, but they do not establish production capacity or reliability. The vendor does not publish service-level guarantees, rate limits, concurrency limits or a support response time. Your LLM provider remains an external dependency, and its availability, quotas and pricing affect the assistant. Check authentication, secret storage, update procedures, logging, backups and what data is sent to each provider before connecting sensitive workflows.

Reddit feedback is mixed and anecdotal. One user described running it on a Raspberry Pi 400 and configuring a Telegram bot; another reported that setup involved workarounds and configuration. A separate commenter said they built it locally for Termux, while another reported easy daily automation. These experiences are not reliability evidence. For production, test your own provider failures, chat integrations, restart behavior and scheduled tasks, and verify that local history and configuration meet your security and compliance requirements.