See me blink
Install Node.js 24, npm, Bash and a C++17 compiler, then run these commands. No model or hardware required.
Full getting-started guideYOUR FAMILY AI AGENT PET
Say Hi Pikoo.
Let’s talk.
I’m Hi Pikoo, a little family AI Agent pet on your desk. Tell me about your day, ask for a bedtime story, or let me help keep track of the little things that matter.
LOCAL WAKE✳LIVE CONVERSATION✳FAMILY MEMORY✳MADE BY YOU
01 / QUICK START
Let’s meet on your screen first.
No board yet? Start with the desktop eye simulator, then build a companion for your desk.
Give me a voice
Configure your own Live model service, Gateway and device credentials to start a real conversation.
Set up voiceBring me to your desk
Add the dual-eye board, a speaker and USB power. Flash the firmware and fit an enclosure you like.
Firmware and installationour first hello
# Local graphics demo · no API key needed
git clone https://github.com/skiyo/pikoo.git
cd pikoo
(cd server && npm ci --ignore-scripts)
bash tools/start-dualeye.shOpen http://127.0.0.1:8793/dualeye. The default is a local graphics and test-tone demo. Real voice needs your own service configuration and microphone permission.
02 / LIFE WITH HI PIKOO
My name is also
how you wake me.
Say “Hi Pikoo” naturally. Once my eyes open and I greet you, we can keep talking.
Say my whole name
Say the full phrase “Hi Pikoo” to wake me from standby.
Wait for my hello
I’ll greet you when the connection is ready. You don’t need to repeat my name for every turn.
Say goodbye when you’re done
Say goodbye to me. Once I finish my farewell, I close my eyes and rest.
In standby, wake-word detection runs on the device. After waking, I connect to the voice service you configure.
CuriousHI PIKOO / DUAL EYE
Oh? What did you find?
Visual demo only. No microphone.
What’s on your mind?
A question, a story, or something that happened today — it all gives us something to talk about.
Keep the little things for next time.
Recent conversations and family memory help us pick up a topic when we meet again.
Need an answer? Let’s find it.
I can help search for information, read web pages, and manage shopping and to-do lists.
Pick an expression and get to know me.
03 / HOW IT WORKS
A place for voice,
thought and memory.
Live handles speech, Terra handles delegated tasks and background work. The Gateway persists text, summaries, family memory and lists. Keys and family profiles stay on your server.
01Dual-eye deviceLocal wake · capture · playback
02GatewayGCP/WSS · auth · persistence
03GPT-Live 1Listen · speak · interrupt
04GPT-6.1 TerraReason · use tools · organize memory
Waveshare ESP32-S3 DualEye board
The device captures audio, plays responses and displays expressions on its two round screens.
Explore the hardwareQwen Audio Agent 2.0.1
Hi Pikoo combines a native Live Provider, Terra delegation, sessions, family memory and tools on this runtime.
View on GitHubContext, memory and lists have distinct roles
Recent text restores topics, digests bound history and long-term memory keeps suitable family facts. Shopping and to-dos live in separate lists.
Read the memory designFind information and handle small tasks
Terra can use configured search, public-page reading and family lists. Add further MCP tools to suit your needs.
Explore tool integration04 / MADE BY YOU
What comes next?
That’s yours to build.
From expressions and enclosures to firmware and the Gateway. Give me new eyes, add a tool, or explore the layer that interests you. Make Hi Pikoo your own.
3D-PRINTED ENCLOSURE
Make a home for these eyes.
The cat face, base and carrier form one main body. Add the cover and rotary lock: just three printed parts. Download the original STLs and print in the order 03 → 01 → 02.
STL files include the main shell, rear cover and rotary lock. Check board, speaker and assembly dimensions before printing.
Read printing & assembly instructions0102030405060708
Device connections
firmware/main/Connect the board, microphone, speaker and dual displays to bring the device together.
Device core
firmware/components/pikoo_core/Manage audio, session state, playback and cancellation on the device.
Expressive eyes
firmware/components/pikoo_eyes/Render the eyes and control gaze, blinking and expression animations.
Desktop simulator
firmware/simulator/Run Hi Pikoo on your computer to refine expressions and interaction flows.
Agent server
server/Compose the Agent runtime and connect realtime models, memory and tools.
Local wake word
training/Prepare, train and evaluate local wake-word models.
Developer tools
tools/Launch, build, flash and diagnose your device.
Documentation
docs/Find architecture, deployment, hardware and development guides.
05 / BEFORE WE MEET
A few things before we meet.
What do I need?
For a physical build, use the Waveshare ESP32-S3-DualEye-LCD-1.28, a compatible speaker, USB power, 2.4GHz Wi-Fi and your own Gateway. The graphics simulator needs no board.
Is this a ready-to-use product?
Hi Pikoo is an open-source DIY project. You provide the hardware, deploy the services and configure them. Start with the graphics demo, which needs no API keys.
Does Hi Pikoo need an internet connection?
Local wake detection does not upload standby audio. Live conversation and online tools require an internet connection. Spoken replies are currently in Chinese by default.
What does it cost to run?
Costs depend on the model, server and external tool services you choose.
Who manages my family’s data?
Memory is managed by the server you deploy; realtime voice also passes through the model service you configure.
How are code and models licensed?
Original code, graphics and documentation use Apache-2.0. Wake weights and model-containing firmware carry separate non-commercial terms and source attribution. Child recall and endurance still need real-world validation.
Your turn to say hello.
Hi Pikoo!