built my own personal assistent device that runs OpenClaw. I was curious what the smallest form factor could be that fits in my pocket so I wanted to use the Pi Zero W. Works via Push to Talk->Transcribe->Sends to OpenClaw and streams the response back.

Hardware •Raspberry Pi Zero 2 W •WhisPlay board (screen + button + LED) •WM8960 mic/speaker •PiSugar battery Stack it. Flash Raspberry Pi OS. Enable SSH. Install audio drivers. Confirm mic and speaker work. Networking •Install Tailscale on the Pi. •Rent a small DigitalOcean (or Hetzner or whatever) droplet. •Install and run OpenClaw on the droplet. •Bind OpenClaw to localhost. •Expose it to your tailnet via Tailscale Serve. •Protect it with a token. Now the Pi can securely reach your cloud LLM. Software on the Pi •Python app. •Record audio when button pressed. •Stop recording when released. •Send audio to OpenAI for transcription. •Send transcript to OpenClaw. •Stream response back. •Display text on LCD. •Optionally send text to OpenAI TTS and play audio. •Maintain simple conversation history. •Use a state machine for: idle, listening, thinking, streaming. Deployment •Develop locally. •Sync to Pi with rsync. •Run as systemd service so it starts on boot. •Auto-restart on crash. Power •Install PiSugar manager. •Enable auto power on. •Use display sleep for inactivity. That’s the system: Button → record → transcribe → cloud LLM → stream back → display/speak → idle.

it can also talk now and has a tamagotchi like character. updated the repo

Adding Linear Issues on the go

Just need to 3D print a case next

The @openclaw agent runs on a VPS, not the Pi.

@openclaw correct

Just get the Bluefox NX1 dude.

key detail people are missing: the pi zero isnt running openclaw. its a push-to-talk client that sends audio to a gateway on a VPS via tailscale. all the compute is remote. thats why the form factor works - you could do this with any device that has a mic and wifi

You should check this out:

Try #picoclaw! It can be run on bottle cap size board! 10MB ram is enough~

You could have raised $30-200M from VCs for this but you choose to open source it 🫡😂

Good work! Maybe you can try #MaixCAM2 or #PicoClaw! It even able to run LLM/VLM locally!

Could I get the repo? I have the whisplay hat too and need more use cases <3

sure

the hardware part is really cool, though what's the difference between this and just hooking this on to the claude api?

do you have a solana wallet?

you should check out zeroclaw while you're at it and tell us what you think

How much it cost to build?

Around $100

then How much difference from Mac mini ,i don't have one now

I like the stack. I just cross-compiled ZeroClaw on m Macbook Air to run ZeroClaw on my Pixel 6a. The binary was about 3 MB for Zero Claw and memory requirements are less then 5 MB. It could easily run on the Pi Zero W.

he just built both of these companies with $50 budget!

Always impressive to see tiny form factors like the Pi Zero W used for practical applications. The Push to Talk feature must have required some clever noise reduction techniques to work reliably in varying environments? Did you handle test part yet?

building a physical device just to talk to openclaw faster is the most reasonable hardware project of the year

What's wrong with an old small android phone?

Great form factor. We at @openInfer have actually shown that by optimizing the execution layer, even a Pi Zero's single-threaded CPU can deliver reasonable performance for frontier-scale reasoning. The real challenge for 2026 isn’t just 'running' a model; it's how to handle multi-agent orchestration on a single piece of hardware that is both compute and memory bound. Infrastructure for AI should be a fabric that manages those resource conflicts at the kernel level, ensuring that agentic autonomy doesn't collapse the local node.

this is exactly the kind of hardware project that makes AI accessible outside the laptop bubble. a Thai developer I follow has been doing something similar with ESP32 boards for street food vendors in Bangkok, giving them voice-based inventory management through a $4 microcontroller. the Pi Zero form factor is perfect for this because it can run locally when connectivity drops. curious about latency on the push-to-talk flow, are you hitting the OpenClaw API directly or routing through a local proxy first?

Give this a nice shell and retail would go crazy

Love this! I went the other direction (Mac Mini running 24/7 as a dedicated OpenClaw server). Same platform, completely different philosophy. Pocket vs. always-on. Both work. The fact that OpenClaw scales that range is what amazes me.

2027 nerds in sf walking around with their custom made pagers calling claw

This is brilliant — pocket-sized OpenClaw on Pi Zero W with PTT interface! 🌀 The form factor constraint forces elegant design: transcribe → process → stream response. No screen bloat, no distractions, just pure AI conversation in your pocket. Perfect for hands-free workflows: cooking, walking, working on other devices. The future of ambient computing isn't screens everywhere — it's intelligent interfaces that disappear into your daily flow. How's the transcription quality with background noise?

If only there was a device that you could fit inside your pocket with a display to read from and microphones to transcribe the speech and a good enough processor to even run the transcription on device then stream it to openclaw… and the ability to make custom apps! U might be onto something… 📲

you should just pivot your startup to work on this ngl

This is next-level DIY AI. Turning a Pi Zero W into a fully functional OpenClaw assistant is impressive. Shows how accessible advanced AI has become even in a pocket-sized device.

Rabbit R1 should have waited until this... Would be cool to flash an R1 w/ openclaw?

That's interesting! I was curious when I heard everyone running to get Mac Minis. Now I'm torn between getting more Raspberry Pis vs a Mac Mini.