Real-time social robotics, from the cloud to your local device.
Watch Ian from our DevX team use Gemini Live for a seamless voice chat with Reachy Mini.
Then, stick around until the end to see the robot running locally on Gemma 4!
We mentioned it a few weeks ago — Reachy Mini's price is changing, for a number of reasons.
You have two days left to get it at the original price: https://t.co/cBP18lVh1Z
Reachy mini running locally in near real time was not on my bingo card too! 😱
With version 1.8.0 you can even add MCP thanks to @ailozovskaya details here: https://t.co/pZ7hKXcaeH (I hope Private Spaces support will be added soon)
AI is speeding up innovation faster than light!
Models used:
- Gemma 4 E4B QAT
- Parakeet TDT 0.6B v3
- Qwen3-TTS 1.7B CustomVoice (speaker: Aiden, you can change this easily)
Commands used:
- llama-server -hf google/gemma-4-E4B-it-qat-q4_0-gguf -np 2 -c 65536 -fa on --swa-full
- speech-to-speech --responses_api_base_url "http://127.0.0.1:8080" --responses_api_api_key ""
Thanks again to @andimarafioti@pollenrobotics@huggingface 🙏
This little robot is so funny to use!
Introducing local Reachy Mini conversations: free chats forever!
So fast that we had to hardcode delays to stop it from interrupting you mid-sentence.
We built an open-source Realtime API powered by llama.cpp:
Parakeet -> Gemma 4 E4B -> Qwen3TTS
Run it anywhere you run local LMs. Video shows DGX Spark and a 36GB M3 Pro MacBook.
Blog: https://t.co/3acRikm2HZ
models are the easiest part of an AI product now
the next phase of AI engineering is better behavior runtimes - everything around the model
when we combine an LLM with robots:
- model sets intent, not direct action
- personality is policy
- idle state is a product decision
- latency is interaction design
- voice is UX
video featuring Reachy Mini from @huggingface x @pollenrobotics x @elevenlabs x @openai on my channel
https://t.co/xsbvoyOcma
also can spot my chat with the lovely @ZhangsqNo1 at @AIEMiami❤️🔥
The panda master is a modified Reachy Mini from @pollenrobotics:
- added panda ears
- painted black eye circles
- connected to GPT voice conversation
When the conversation reaches “fortune telling mode,” GPT emits a tool call containing:
- drawing trajectories
- interpretation text
While generating the fortune map, the panda’s head starts shaking violently like it’s performing some ancient ritual 😂
BTS 🤖
Team: My kids + nephews joined as interns. Lots of questions. One 2-year-old QA tester nearly started a fish tank test 📷
Thanks to @huggingface and @pollenrobotics
to make AI and robotics feel real, hands-on, and exciting for the next generation of builders.