PicoClaw Review

I installed PicoClaw on a Mac. It was freshly set up with a new user and does not have access to my iCloud data or emails. The special feature of my setup is that I also run the AI model on the Mac and PicoClaw is not run with OpenAI or Claude. The installation and setup was very easy:

• brew install llama.cpp picoclaw And just like that, I had everything I needed (ollama can also be used instead of llama.cpp).

I started with the GLM-4.7 Flash (4-bit quantized) AI model. Today, I use the Qwen3.5 Flash AI model. Both work perfectly for my use cases. In its “out-of-the-box” state, PicoClaw has the ability to read, write, or edit files in its workspace. It can also execute commands. This allows it to access websites and other things.

The most exciting and important part is the skills. Skills describe abilities for AI agents (such as the use of certain weather APIs). Since skills are universal, they can be exchanged between agent systems such as OpenClaw, NanoBot, Claude Code, etc. I have installed skills for openhue, blogwatcher, code review, SEO, copywriting, etc. in my PicoClaw.

Using these skills, I was able to successfully complete the following tasks with PicoClaw:

• Turning my Philips Hue lights on and off
• Finding and reading new blog articles
• Determining the opening hours of local stores
• Weather forecast for my city
• Summary of my email inbox
• Programming small programs (vocabulary trainer for my children)
• Adding, removing, and listing items on a shopping list.
• Remembering to-dos (drink a glass of water every 2 hours).
• Perform code review
• Perform website reviews

Communication took place via Telegram Messenger. I placed my emails in the workspace as mbox files using an IMAP backup script (security first). Since I had already worked extensively with opencode (agentic coding) before, most of it was not new to me. However, I was impressed by its memory capabilities. Having an assistant perform regular tasks and then receiving the results expands my use of AI.

I highly recommend everyone give PicoClaw a try.

To enable all these skills locally on my MacBook, I had to set up a few things:

1. LLM download and start

Install Llama.cpp

brew install llama.cpp

Start llama-server with GLM-4.7-Flash

llama-server \
    --model GLM-4.7-Flash-UD-Q4_K_XL.gguf \
    --alias "unsloth/GLM-4.7-Flash" \
    --fit on \
    --temp 1.0 \
    --top-p 0.95 \
    --min-p 0.01 \
    --ctx-size 16384 \
    --port 8001 \
    --jinja

2. install and start PicoClaw

Install PicoClaw

brew install picoclaw

init PicoClaw

picoclaw onboard

configurate PicoClaw (~/.picoclaw/config.json):

{
  "agents": {
    "defaults": {
      "workspace": "~/.picoclaw/workspace",
      "restrict_to_workspace": true,
      "provider": "openai",
      "model": "unsloth/GLM-4.7-Flash",
      "max_tokens": 202752,
      "temperature": 1.0,
      "max_tool_iterations": 20
    }
  },
  "channels": {
    "telegram": {
      "enabled": true,
      "token": "botfather-token-einfügen",
      "proxy": "",
      "allow_from": []
    }
  },
  "providers": {
    "openai": {
      "api_key": "egal",
      "api_base": "http://localhost:8001/v1"
    }
  },
  "gateway": {
    "host": "0.0.0.0",
    "port": 18790
  },
  "tools": {
    "web": {
      "brave": {
        "enabled": false,
        "api_key": "",
        "max_results": 5
      },
      "duckduckgo": {
        "enabled": true,
        "max_results": 5
      }
    }
  },
  "heartbeat": {
    "enabled": true,
    "interval": 30
  },
  "devices": {
    "enabled": false,
    "monitor_usb": true
  }
}

3. Add skills

I removed the Summerize skill because it didn’t work for me:

~/.picoclaw/workspace/skills/summarize

I added the skills openhue and blogwatcher:

• https://skills.sh/steipete/clawdis/openhue
• https://skills.sh/steipete/clawdis/blogwatcher

Für beide Skills habe ich die Single-Go-Binaries geladen:

• https://github.com/openhue/openhue-cli
• https://github.com/Hyaxia/blogwatcher

After connecting openhue to my Hue and adding my favorite blogs to blogwatcher, I was finally ready to start. Communication ran via Telegram, which I connected via Botfather (put the token in config.json).

4. Reading my email inbox

To enable PicoClaw to access my inbox, I created a cron job that saves my IMAP inbox to the PicoClaw workspace. This allows PicoClaw to access my inbox without passwords. I used a Python script for this:

https://github.com/rcarmo/imapbackup/blob/master/imapbackup312.py

python imapbackup312.py \
  --server imap.example.com:993 \
  --ssl \
  --user your.name@example.com \
  --pass @/path/to/secret.txt \
  --mbox-dir ~/.picoclaw/workspace/mail_backup \
  --folders INBOX,Sent,Archive \
  --yes-overwrite-mboxes \
  --quiet