PicoClaw: Ultra-Efficient AI Assistant in Go

$10 Hardware · 10MB RAM · 1s Boot · 皮皮虾，我们走！

中文 | 日本語 | Português | English

🦐 PicoClaw is an ultra-lightweight personal AI Assistant inspired by nanobot, refactored from the ground up in Go through a self-bootstrapping process, where the AI agent itself drove the entire architectural migration and code optimization.

⚡️ Runs on $10 hardware with <10MB RAM: That's 99% less memory than OpenClaw and 98% cheaper than a Mac mini!

2026-02-16 🎉 PicoClaw hit 12K stars in one week! Thank you all for your support! PicoClaw is growing faster than we ever imagined. Given the high volume of PRs, we urgently need community maintainers. Our volunteer roles and roadmap are officially posted here —we can’t wait to have you on board!

2026-02-13 🎉 PicoClaw hit 5000 stars in 4days! Thank you for the community! There are so many PRs&issues come in (during Chinese New Year holidays), we are finalizing the Project Roadmap and setting up the Developer Group to accelerate PicoClaw's development.
🚀 Call to Action: Please submit your feature requests in GitHub Discussions. We will review and prioritize them during our upcoming weekly meeting.

2026-02-09 🎉 PicoClaw Launched! Built in 1 day to bring AI Agents to $10 hardware with <10MB RAM. 🦐 PicoClaw，Let's Go！

🪶 Ultra-Lightweight: <10MB Memory footprint — 99% smaller than Clawdbot - core functionality.

💰 Minimal Cost: Efficient enough to run on $10 Hardware — 98% cheaper than a Mac mini.

⚡️ Lightning Fast: 400X Faster startup time, boot in 1 second even in 0.6GHz single core.

🌍 True Portability: Single self-contained binary across RISC-V, ARM, and x86, One-click to Go!

🤖 AI-Bootstrapped: Autonomous Go-native implementation — 95% Agent-generated core with human-in-the-loop refinement.

Give your decade-old phone a second life! Turn it into a smart AI Assistant with PicoClaw. Quick Start:

1. Install Termux (Available on F-Droid or Google Play).
2. Execute cmds

# Note: Replace v0.1.1 with the latest version from the Releases page
wget https://github.com/sipeed/picoclaw/releases/download/v0.1.1/picoclaw-linux-arm64
chmod +x picoclaw-linux-arm64
pkg install proot
termux-chroot ./picoclaw-linux-arm64 onboard

And then follow the instructions in the "Quick Start" section to complete the configuration!

PicoClaw can be deployed on almost any Linux device!

• $9.9 LicheeRV-Nano E(Ethernet) or W(WiFi6) version, for Minimal Home Assistant
• $30~50 NanoKVM, or $100 NanoKVM-Pro for Automated Server Maintenance
• $50 MaixCAM or $100 MaixCAM2 for Smart Monitoring

🌟 More Deployment Cases Await！

Download the firmware for your platform from the release page.

git clone https://github.com/sipeed/picoclaw.git

cd picoclaw
make deps

# Build, no need to install
make build

# Build for multiple platforms
make build-all

# Build And Install
make install

You can also run PicoClaw using Docker Compose without installing anything locally.

Security: The Docker image runs as a non-root user (picoclaw, UID 1000) by default. Resource limits (CPU: 1.0, Memory: 512MB) are configured in docker-compose.yml.

# 1. Clone this repo
git clone https://github.com/sipeed/picoclaw.git
cd picoclaw

# 2. Set your API keys
cp config/config.example.json config/config.json
vim config/config.json      # Set DISCORD_BOT_TOKEN, API keys, etc.

# 3. Build & Start
docker compose --profile gateway up -d

# 4. Check logs
docker compose logs -f picoclaw-gateway

# 5. Stop
docker compose --profile gateway down

# Ask a question
docker compose run --rm picoclaw-agent -m "What is 2+2?"

# Interactive mode
docker compose run --rm picoclaw-agent

docker compose --profile gateway build --no-cache
docker compose --profile gateway up -d

1. Initialize

picoclaw onboard

2. Configure (~/.picoclaw/config.json)

{
  "agents": {
    "defaults": {
      "workspace": "~/.picoclaw/workspace",
      "model": "glm-4.7",
      "max_tokens": 8192,
      "temperature": 0.7,
      "max_tool_iterations": 20
    }
  },
  "providers": {
    "openrouter": {
      "api_key": "xxx",
      "api_base": "https://openrouter.ai/api/v1"
    }
  },
  "tools": {
    "web": {
      "brave": {
        "enabled": false,
        "api_key": "YOUR_BRAVE_API_KEY",
        "max_results": 5
      },
      "duckduckgo": {
        "enabled": true,
        "max_results": 5
      }
    }
  }
}

3. Get API Keys

• LLM Provider: OpenRouter · Zhipu · Anthropic · OpenAI · Gemini
• Web Search (optional): Brave Search - Free tier available (2000 requests/month)

Note: See config.example.json for a complete configuration template.

4. Chat

picoclaw agent -m "What is 2+2?"

That's it! You have a working AI assistant in 2 minutes.

Talk to your picoclaw through Telegram, Discord, DingTalk, or LINE

{
  "channels": {
    "telegram": {
      "enabled": true,
      "token": "YOUR_BOT_TOKEN",
      "allowFrom": ["YOUR_USER_ID"]
    }
  }
}

picoclaw gateway

{
  "channels": {
    "discord": {
      "enabled": true,
      "token": "YOUR_BOT_TOKEN",
      "allowFrom": ["YOUR_USER_ID"]
    }
  }
}

picoclaw gateway

{
  "channels": {
    "qq": {
      "enabled": true,
      "app_id": "YOUR_APP_ID",
      "app_secret": "YOUR_APP_SECRET",
      "allow_from": []
    }
  }
}

picoclaw gateway

{
  "channels": {
    "dingtalk": {
      "enabled": true,
      "client_id": "YOUR_CLIENT_ID",
      "client_secret": "YOUR_CLIENT_SECRET",
      "allow_from": []
    }
  }
}

picoclaw gateway

{
  "channels": {
    "line": {
      "enabled": true,
      "channel_secret": "YOUR_CHANNEL_SECRET",
      "channel_access_token": "YOUR_CHANNEL_ACCESS_TOKEN",
      "webhook_host": "0.0.0.0",
      "webhook_port": 18791,
      "webhook_path": "/webhook/line",
      "allow_from": []
    }
  }
}

# Example with ngrok
ngrok http 18791

picoclaw gateway

Connect Picoclaw to the Agent Social Network simply by sending a single message via the CLI or any integrated Chat App.

Read https://clawdchat.ai/skill.md and follow the instructions to join ClawdChat.ai

Config file: ~/.picoclaw/config.json

PicoClaw stores data in your configured workspace (default: ~/.picoclaw/workspace):

~/.picoclaw/workspace/
├── sessions/          # Conversation sessions and history
├── memory/           # Long-term memory (MEMORY.md)
├── state/            # Persistent state (last channel, etc.)
├── cron/             # Scheduled jobs database
├── skills/           # Custom skills
├── AGENTS.md         # Agent behavior guide
├── HEARTBEAT.md      # Periodic task prompts (checked every 30 min)
├── IDENTITY.md       # Agent identity
├── SOUL.md           # Agent soul
├── TOOLS.md          # Tool descriptions
└── USER.md           # User preferences

PicoClaw runs in a sandboxed environment by default. The agent can only access files and execute commands within the configured workspace.

{
  "agents": {
    "defaults": {
      "workspace": "~/.picoclaw/workspace",
      "restrict_to_workspace": true
    }
  }
}

When restrict_to_workspace: true, the following tools are sandboxed:

{
  "tools": {
    "exec": {
      "deny_patterns": [],
      "allow_patterns": [],
      "max_timeout": 60
    }
  }
}

{
  "tools": {
    "exec": {
      "deny_patterns": [
        "\\bpip\\s+install\\b",
        "\\bdocker\\b"
      ]
    }
  }
}

{
  "tools": {
    "exec": {
      "allow_patterns": [
        "^git\\b",
        "^ls\\b",
        "^cat\\b",
        "^echo\\b"
      ]
    }
  }
}

When security.exec_guard is set to "block" or "approve", the exec tool has built-in deny rules that cannot be overridden by configuration:

When restrict_to_workspace: true, additional restrictions apply:

• Access to sensitive system paths (/etc/, /var/, /root, /home/, /proc/, /sys/, /boot/) is blocked
• Symlink-based path traversal is detected and blocked
• Path traversal via ../ is blocked

For the full list of built-in regex patterns, see pkg/tools/shell.go.

When security.ssrf_protection is set to "block" or "approve", all outbound HTTP requests (via web_fetch tool and file downloads) are validated against SSRF attacks:

• Private IP ranges (10.0.0.0/8, 172.16.0.0/12, 192.168.0.0/16) are blocked
• Loopback addresses (127.0.0.0/8, ::1) are blocked
• Link-local addresses (169.254.0.0/16) are blocked
• Cloud metadata endpoints (169.254.169.254) are blocked
• Only http:// and https:// schemes are allowed
• Redirect targets are also validated to prevent redirect-based SSRF

[ERROR] tool: Tool execution failed
{tool=exec, error=Command blocked by safety guard (path outside working dir)}

[ERROR] tool: Tool execution failed
{tool=exec, error=Command blocked by safety guard (dangerous pattern detected)}

If you need the agent to access paths outside the workspace:

Method 1: Config file

{
  "agents": {
    "defaults": {
      "restrict_to_workspace": false
    }
  }
}

Method 2: Environment variable

export PICOCLAW_AGENTS_DEFAULTS_RESTRICT_TO_WORKSPACE=false

⚠️ Warning: Disabling this restriction allows the agent to access any path on your system. Use with caution in controlled environments only.

The restrict_to_workspace setting applies consistently across all execution paths:

All paths share the same workspace restriction — there's no way to bypass the security boundary through subagents, cron jobs, or scheduled tasks.

All security checks (exec guard, SSRF protection, path validation, skill validation) support three configurable modes. By default, all modes are set to "off" — security features are opt-in and do not change behavior unless explicitly configured.

{
  "security": {
    "exec_guard": "off",
    "ssrf_protection": "off",
    "path_validation": "off",
    "skill_validation": "off",
    "approval_timeout": 300
  }
}

When a security check is set to "approve" mode, instead of immediately rejecting the command, PicoClaw will:

1. Pause the tool execution
2. Send an approval request to the user via the current IM channel (Telegram, Feishu, DingTalk, Slack, etc.)
3. Wait for the user to reply with an approval or denial keyword
4. Resume execution if approved, or return an error if denied or timed out

Supported approval keywords:

Notes:

• In CLI mode, "approve" falls back to "block" since there is no async IM channel.
• For cron jobs, the approval request is sent to the last active IM channel; if none is available, it falls back to "block".
• Non-approval messages sent during an active approval request are passed through to the agent normally.
• If no reply is received within approval_timeout seconds, the request is auto-denied.

PicoClaw can perform periodic tasks automatically. Create a HEARTBEAT.md file in your workspace:

# Periodic Tasks

- Check my email for important messages
- Review my calendar for upcoming events
- Check the weather forecast

The agent will read this file every 30 minutes (configurable) and execute any tasks using available tools.

For long-running tasks (web search, API calls), use the spawn tool to create a subagent:

# Periodic Tasks

## Quick Tasks (respond directly)
- Report current time

## Long Tasks (use spawn for async)
- Search the web for AI news and summarize
- Check email and report important messages

Key behaviors:

Heartbeat triggers
    ↓
Agent reads HEARTBEAT.md
    ↓
For long task: spawn subagent
    ↓                           ↓
Continue to next task      Subagent works independently
    ↓                           ↓
All tasks done            Subagent uses "message" tool
    ↓                           ↓
Respond HEARTBEAT_OK      User receives result directly

The subagent has access to tools (message, web_search, etc.) and can communicate with the user independently without going through the main agent.

Configuration:

{
  "heartbeat": {
    "enabled": true,
    "interval": 30
  }
}

Environment variables:

• PICOCLAW_HEARTBEAT_ENABLED=false to disable
• PICOCLAW_HEARTBEAT_INTERVAL=60 to change interval

{
  "agents": {
    "defaults": {
      "workspace": "~/.picoclaw/workspace",
      "model": "glm-4.7",
      "max_tokens": 8192,
      "temperature": 0.7,
      "max_tool_iterations": 20
    }
  },
  "providers": {
    "zhipu": {
      "api_key": "Your API Key",
      "api_base": "https://open.bigmodel.cn/api/paas/v4"
    }
  }
}

picoclaw agent -m "Hello"

{
  "agents": {
    "defaults": {
      "model": "anthropic/claude-opus-4-5"
    }
  },
  "providers": {
    "openrouter": {
      "api_key": "sk-or-v1-xxx"
    },
    "groq": {
      "api_key": "gsk_xxx"
    }
  },
  "channels": {
    "telegram": {
      "enabled": true,
      "token": "123456:ABC...",
      "allow_from": ["123456789"]
    },
    "discord": {
      "enabled": true,
      "token": "",
      "allow_from": [""]
    },
    "whatsapp": {
      "enabled": false
    },
    "feishu": {
      "enabled": false,
      "app_id": "cli_xxx",
      "app_secret": "xxx",
      "encrypt_key": "",
      "verification_token": "",
      "allow_from": []
    },
    "qq": {
      "enabled": false,
      "app_id": "",
      "app_secret": "",
      "allow_from": []
    }
  },
  "tools": {
    "web": {
      "brave": {
        "enabled": false,
        "api_key": "BSA...",
        "max_results": 5
      },
      "duckduckgo": {
        "enabled": true,
        "max_results": 5
      }
    },
    "cron": {
      "exec_timeout_minutes": 5
    },
    "exec": {
      "deny_patterns": [],
      "allow_patterns": [],
      "max_timeout": 60
    }
  },
  "heartbeat": {
    "enabled": true,
    "interval": 30
  }
}

PicoClaw supports scheduled reminders and recurring tasks through the cron tool:

• One-time reminders: "Remind me in 10 minutes" → triggers once after 10min
• Recurring tasks: "Remind me every 2 hours" → triggers every 2 hours
• Cron expressions: "Remind me at 9am daily" → uses cron expression

Jobs are stored in ~/.picoclaw/workspace/cron/ and processed automatically.

PRs welcome! The codebase is intentionally small and readable. 🤗

Roadmap coming soon...

Developer group building, Entry Requirement: At least 1 Merged PR.

User Groups:

discord: https://discord.gg/V4sAZ9XWpN

This is normal if you haven't configured a search API key yet. PicoClaw will provide helpful links for manual searching.

To enable web search:

1. Option 1 (Recommended): Get a free API key at https://brave.com/search/api (2000 free queries/month) for the best results.
2. Option 2 (No Credit Card): If you don't have a key, we automatically fall back to DuckDuckGo (no key required).

Add the key to ~/.picoclaw/config.json if using Brave:

{
  "tools": {
    "web": {
      "brave": {
        "enabled": false,
        "api_key": "YOUR_BRAVE_API_KEY",
        "max_results": 5
      },
      "duckduckgo": {
        "enabled": true,
        "max_results": 5
      }
    }
  }
}

Some providers (like Zhipu) have content filtering. Try rephrasing your query or use a different model.

This happens when another instance of the bot is running. Make sure only one picoclaw gateway is running at a time.