Skip to main content

General Questions

PicoClaw is an ultra-lightweight personal AI assistant inspired by nanobot, refactored from the ground up in Go. It runs on $10 hardware with <10MB RAM, making it 99% less memory-hungry than OpenClaw and 98% cheaper than a Mac mini.Key features:
  • Ultra-lightweight (<10MB memory footprint)
  • Lightning fast (400X faster startup, boots in 1 second)
  • True portability (single binary across RISC-V, ARM, and x86)
  • AI-bootstrapped (95% agent-generated core)
FeatureOpenClawNanoBotPicoClaw
LanguageTypeScriptPythonGo
RAM>1GB>100MB< 10MB
Startup (0.8GHz)>500s>30s<1s
CostMac Mini $599Linux SBC ~$50Any Linux Board ~$10
PicoClaw is designed for extreme resource efficiency, making it ideal for embedded devices and edge computing.
PicoClaw can run on:
  • Embedded Boards: LicheeRV-Nano (9.9),NanoKVM(9.9), NanoKVM (30-100), MaixCAM ($50-100)
  • Single Board Computers: Raspberry Pi (including Pi Zero 2 W)
  • Old Android Phones: Via Termux
  • Servers & Desktops: Any Linux, macOS, or Windows machine
  • Docker Containers: Portable deployment
Minimum requirements:
  • 64MB RAM (target: <20MB)
  • Any architecture: x86_64, ARM64, ARM32, RISC-V
Yes! PicoClaw is open source under the MIT license. You can:
  • View the source code on GitHub
  • Contribute features, bug fixes, or documentation
  • Fork and customize for your needs
  • Use it commercially or personally
See our Contributing Guide to get started.
Yes! PicoClaw was substantially developed with AI assistance through a self-bootstrapping process. The AI agent itself drove the architectural migration from Python to Go and code optimization.However, this doesn’t mean the code is unreviewed or low-quality:
  • All AI-generated code is reviewed and validated by humans
  • Contributors are responsible for understanding and testing code
  • The same quality standards apply to all contributions
We embrace responsible AI-assisted development and have built our contribution process around it.

Setup Questions

Download from the releases page:
# Linux/macOS
chmod +x picoclaw-*
./picoclaw-* onboard
Required:
  • LLM Provider: Choose one:
    • OpenRouter - Access to all models (200K tokens/month free)
    • Zhipu - Best for Chinese users (200K tokens/month free)
    • OpenAI - GPT models
    • Anthropic - Claude models
    • Gemini - Google models
Optional:
  • Web Search: Choose one:
    • Brave Search - 2000 queries/month free
    • Tavily - 1000 queries/month free, AI-optimized
    • DuckDuckGo - Built-in fallback, no key needed
  • Voice Transcription: Groq - Free Whisper transcription
After running picoclaw onboard, edit ~/.picoclaw/config.json:Basic Setup:
{
  "model_list": [
    {
      "model_name": "gpt4",
      "model": "openai/gpt-5.2",
      "api_key": "your-api-key"
    }
  ],
  "agents": {
    "defaults": {
      "model": "gpt4"
    }
  }
}
See config.example.json for a complete template.
Download the latest release and replace your existing binary:
wget https://github.com/sipeed/picoclaw/releases/latest/download/picoclaw-linux-amd64
chmod +x picoclaw-linux-amd64

Usage Questions

Command Line:
# One-shot message
picoclaw agent -m "What is 2+2?"

# Interactive mode
picoclaw agent
Chat Apps (after configuring channels):
picoclaw gateway
# Then message your bot on Telegram, Discord, WhatsApp, etc.
PicoClaw is a full-featured AI assistant that can:File Operations:
  • Read, write, edit files
  • List directories
  • Search file contents
Web & Research:
  • Search the web
  • Fetch and analyze web pages
  • Stay updated on news and trends
Development:
  • Write and debug code
  • Execute shell commands
  • Manage git repositories
Planning & Memory:
  • Create and manage schedules
  • Set reminders and recurring tasks
  • Remember important information across sessions
Communication:
  • Respond on multiple chat platforms
  • Send messages to users
  • Handle group chats
See our detailed channel guides:Quick Telegram Setup:
  1. Message @BotFather on Telegram
  2. Send /newbot and follow prompts
  3. Copy the token to config.json:
{
  "channels": {
    "telegram": {
      "enabled": true,
      "token": "YOUR_BOT_TOKEN",
      "allow_from": ["YOUR_USER_ID"]
    }
  }
}
  1. Run picoclaw gateway
Yes! PicoClaw supports two types of scheduling:1. Heartbeat Tasks (periodic): Create ~/.picoclaw/workspace/HEARTBEAT.md:
# Periodic Tasks

- Check my email for important messages
- Review my calendar for upcoming events
- Search for AI news and summarize
The agent checks this file every 30 minutes (configurable).2. Cron Jobs (one-time or recurring):
# Via CLI
picoclaw cron add "Remind me to call mom" "in 1 hour"
picoclaw cron list

# Via chat
"Remind me in 10 minutes to check the oven"
"Remind me every day at 9am to review my schedule"
PicoClaw has multiple memory mechanisms:Session Memory:
  • Stored in ~/.picoclaw/workspace/sessions/
  • Preserves conversation context
  • Separate sessions per channel/user
Long-term Memory:
  • MEMORY.md - Important facts and preferences
  • USER.md - User profile and preferences
  • IDENTITY.md - Agent identity and personality
State Persistence:
  • state/ - Last used channel, active sessions
  • cron/ - Scheduled tasks database
The agent automatically updates these files as it learns about you.
PicoClaw has multiple security features:Sandbox by Default:
  • restrict_to_workspace: true limits file access to workspace only
  • Dangerous commands (rm -rf, format, etc.) are blocked
  • No access to system files outside workspace
Channel Security:
  • allow_from restricts who can use your bot
  • Per-channel session isolation
  • No data leakage between users
API Key Protection:
  • Stored in local config file
  • Never logged or transmitted except to provider
  • Credentials redacted from logs
See our Roadmap for upcoming security enhancements.

Provider Questions

Best for Most Users: OpenRouter
  • Access to all major models (GPT, Claude, Gemini, Llama, etc.)
  • 200K tokens/month free tier
  • Easy switching between models
Best for Chinese Users: Zhipu (智谱)
  • Optimized for Chinese language
  • 200K tokens/month free tier
  • Fast response times in China
Best for Privacy: Local models via Ollama
  • Runs entirely on your hardware
  • No API keys needed
  • No data leaves your machine
Best for Speed: Groq or Cerebras
  • Ultra-fast inference
  • Free tier available
  • Great for simple tasks
Yes! PicoClaw supports local inference via:Ollama (Recommended):
# Install Ollama
curl -fsSL https://ollama.com/install.sh | sh

# Pull a model
ollama pull llama3

# Configure PicoClaw
{
  "model_list": [
    {
      "model_name": "local-llama",
      "model": "ollama/llama3"
    }
  ]
}
Other Options:
  • vLLM
  • LM Studio
  • LiteLLM Proxy
See Provider Configuration for details.
Configure multiple models in model_list:
{
  "model_list": [
    {
      "model_name": "gpt4",
      "model": "openai/gpt-5.2",
      "api_key": "sk-..."
    },
    {
      "model_name": "claude",
      "model": "anthropic/claude-sonnet-4.6",
      "api_key": "sk-ant-..."
    },
    {
      "model_name": "local",
      "model": "ollama/llama3"
    }
  ],
  "agents": {
    "defaults": {
      "model": "gpt4"  // Default model
    }
  }
}
Change the default model in agents.defaults.model to switch.
Load Balancing: Configure multiple endpoints for failover:
{
  "model_list": [
    {
      "model_name": "gpt4",
      "model": "openai/gpt-5.2",
      "api_base": "https://api1.example.com/v1",
      "api_key": "key1"
    },
    {
      "model_name": "gpt4",
      "model": "openai/gpt-5.2",
      "api_base": "https://api2.example.com/v1",
      "api_key": "key2"
    }
  ]
}
PicoClaw will automatically round-robin between endpoints.
Hardware Costs:
  • Embedded board: $10-50 (one-time)
  • Or use existing hardware: $0
API Costs (monthly):
  • OpenRouter: 200K tokens free, then ~$0.50-2/million tokens
  • Zhipu: 200K tokens free, then similar pricing
  • Local models (Ollama): $0, but uses your compute
  • Brave Search: 2000 queries free, then $5/1000 queries
Typical Usage:
  • Light user (<100 messages/day): Usually stays in free tier
  • Heavy user (>500 messages/day): ~$5-15/month
See our API Key Comparison.

Advanced Questions

Yes! Edit these files in your workspace:IDENTITY.md - Agent identity and name:
You are PicoClaw, a helpful AI assistant.
You are efficient, accurate, and friendly.
SOUL.md - Personality and behavior:
You respond concisely and professionally.
You prioritize user needs and ask clarifying questions.
AGENTS.md - Agent behavior guide:
When writing code, always include comments.
When searching, summarize findings concisely.
The agent reads these files on startup and follows the guidelines.
Create skill files in ~/.picoclaw/workspace/skills/:Example: email_checker.json
{
  "name": "check_email",
  "description": "Check email for important messages",
  "parameters": {
    "type": "object",
    "properties": {
      "filter": {
        "type": "string",
        "description": "Filter criteria (e.g., 'from:[email protected]')"
      }
    }
  },
  "handler": "skills/email_checker.sh"
}
PicoClaw will automatically discover and use the skill.
Yes! Each instance needs:Separate Config:
export PICOCLAW_CONFIG=/path/to/instance1/config.json
picoclaw gateway
Separate Workspace:
{
  "agents": {
    "defaults": {
      "workspace": "~/.picoclaw/workspace1"
    }
  }
}
Different Gateway Ports (if using webhooks):
{
  "gateway": {
    "host": "127.0.0.1",
    "port": 18790  // Use different port for each instance
  }
}
See our Roadmap for upcoming “Swarm Mode” (multi-agent collaboration).
We welcome all contributions! See our Contributing Guide.Quick Start:
  1. Fork the repository
  2. Create a feature branch
  3. Make your changes
  4. Run make check to verify
  5. Open a Pull Request
What to Contribute:
  • Bug fixes
  • New features
  • Documentation improvements
  • New channel integrations
  • Provider support
  • Testing on new hardware
Community Channels:
See our full Roadmap for details. Highlights:Near-term:
  • Memory footprint reduction (<20MB target)
  • Security hardening (sandboxing, SSRF protection)
  • Protocol-first provider architecture
  • More channel integrations
Mid-term:
  • MCP (Model Context Protocol) support
  • Browser automation
  • Mobile device control
  • Multi-agent collaboration
Long-term:
  • Swarm mode (multiple PicoClaw instances)
  • Model routing (smart task distribution)
  • AIEOS (AI-native OS interaction)

Still Have Questions?

Troubleshooting

Common issues and solutions

GitHub Issues

Report bugs or request features

GitHub Discussions

Ask questions and share ideas

Discord

Join the community chat

Build docs developers (and LLMs) love