Pierre-yves Baloche (@FunkyPiwy) header
Pierre-yves Baloche (@FunkyPiwy) profile picture
Joined February 2011
✋ Local models don't make you safe: your agent can read your env vars and your whole home directory. Mini Me v1.0.5 brings mm dockerinit: a Dockerfile + compose.yaml, and the agent runs in a container, as a non-root user. mini-me.org/p/20261107-mm-se… #Docker #LocalModels
1
2
4
389
Pierre-yves Baloche retweeted
In case you missed it, this is my StackChan Nous Girl Tribute edition - running Hermes-Agent for the brains and StackChan Robot for the body. All open source to use and play with. Use any local, sub or api for LLM and voice models. REPO UPDATE: The repo has also been updated to include Nous Girl Robot SOUL.MD so you can learn more about persona development and other interesting tid bits about creating a personality for your robot. Its under Theory of mind: uncertainty and misunderstanding repair on readme/homepage. The correct repo link: github.com/dax8it/nous-girl-… @M5Stack @NousResearch
This is my Stackchan Tribute of Nous-Girl, a little desk robot I've been building for fun. She talks, listens, makes faces, dances, plays music, does DJ and rave mode, shows memes, and can even send email. Repo: github.com/nous-girl-stackch… I'm not an expert at this, but tried my best to write this guide, it's a work in progress. Will keep adding to it as I have time! Adding more features hopefully everyday. What you need: an M5Stack StackChan (CoreS3), a USB-C cable and a computer (Windows, Mac or Linux). For memory, intelligence and extra skills, you will need Hermes-Agent from @NousResearch. The easiest way to start: Download a beta build with a starter face (robot, boy or girl). Flash it from your browser or with one command. On first boot, point it at the official xiaozhi cloud and start talking. No server needed. Want more privacy and control? Run your own server for the fancy stuff, like DJ mode, memes, email and adding a smarter Hermes Agent brain. Every feature has step-by-step docs. Heads up: this is beta. I've only tested on my own robot, so things may break. Please report issues on GitHub and I'll help the best I can. Coming next: simpler builds that need less setup and less upkeep, a version that runs fully from the Nous Research cloud, more languages and local models. Repo: github.com/nous-girl-stackch… (StackChan + Hermes Agent Bot) @M5Stack StackChan Robot @agentmail Robot EMail @GradiumAI Voice TTS @ollama Local LLM @bot Hermes partner in crime
3
3
2
19
1,518
Pierre-yves Baloche retweeted
it figured out the wiring itself, all I did was wiring up the parts in 2d wiring workspace and it figure out the rest. kevsrobots.com/projects/hub
2
4
55
2,247
En pleine préparation d'un tuto vidéo sur le @Creality3dP #3Dscanner #SermoonP1S ! 🎥📐 Du scan brut (laser/IR) jusqu'à l'export du modèle final pour l'impression 3D (STL/3MF), je vous explique tout le processus étape par étape. Bientôt disponible sur YouTube ! #Creality #3DScan
1
4
270
Pierre-yves Baloche retweeted
The Instruction Steps interface; you drop in your STL files, connect them with joins and say which parts to add per step
2
4
50
2,296
Pierre-yves Baloche retweeted
I've made a tool for creating instruction manuals for projects. You can connect the wires in a 2d wiring diagram, and then it will route the wires in 3d.
1
4
39
1,170
Pierre-yves Baloche retweeted
I've upgraded the instruction builder on the kevsrobots community projects page - you can now create Ikea/Lego like instruction guides just using the STL files you already have.
4
6
61
2,930
Pierre-yves Baloche retweeted
Start to ship the first batch of CardputerZero.
39
65
20
449
32,794
Pierre-yves Baloche retweeted
Rooting an Amazon Echo Dot blog.adafruit.com/2026/11/02…
11
2
102
9,141
I wanted a Slack for my little local agents, not another CLI. So I built Sidekicks: channels, @ mentions, plain-words permissions, and any #ACP agent (Claude, Codex...) plugged in, even in @Docker sbx sandboxes. sidekicks.world/p/20261102-s… #AI #LocalModels
5
1
10
665
Pierre-yves Baloche retweeted
Meet Qwen3.8-27B-SLQ-GGUF: a quantized version of Qwen3.8-27B that runs efficiently on llama.cpp. With SLQ mixed-precision, it balances quality and speed for local text generation. 9k+ downloads and counting!
4
2
28
2,503
Pierre-yves Baloche retweeted
My open duck is nearly printed out. Just some large prints left to do!
6
4
105
4,614
Pierre-yves Baloche retweeted
I've released Projects 2.0 on kevsrobots - the best place to document your robotics projects. Add parts to a bill of materials from a catalog, create wiring diagrams, all in standard Markdown. Its easy and its free! kevsrobots.com/projects/kev/…
1
3
14
681
Pierre-yves Baloche retweeted
⌨️ 太卷了,OpenIDEA 照着 IntelliJ IDEA 的手感重做了一个 IDE,打开项目 0.9 秒,内存只占 169MB。 OpenIDEA 的仓库 10 月 21 日才建,10 月 28、29 两天就连发了四个版本,issue 区已经有六十多个开着。 用过 IDEA 的都懂:项目一大,开机等索引、风扇狂转,内存两个 G 起步。 OpenIDEA 用 Rust 做内核,键位、工具窗口、全局搜索都沿用 IDEA 的习惯,肌肉记忆不用改;编辑、调试、Git、终端、数据库和 HTTP 请求也都在一个窗口里。 项目拿 freqtrade 这个 Python 项目做了对比:从打开到 CPU 平稳,OpenIDEA 1.1 秒,IDEA 13.2 秒;内存 169MB 对 2011MB。文档也老实写了,IDEA 对 Python 的分析更深,而且项目和 JetBrains 没有任何关系。 熟悉的手感,轻得多的身子。 GitHub:github.com/l0ng-ai/openidea
17
41
6
322
37,167
Nothing forces an ACP client to be a code editor. Any program that can launch a process can drive Mini Me "as a library". Node.js + the ACP SDK: stream answers, allow or refuse tools, cancel. No server, no port: two pipes. mini-me.org/p/20261101-mm-as… #ACP #LocalModels
2
1
2
474
Pierre-yves Baloche retweeted
Next Wednesday
27
52
13
498
48,642
Aujourd'hui pour mes ... 73 ans, un grand merci à toutes celles et ceux qui y ont pensé ! Et pour tous les radioamateurs et passionnés d'ondes qui me lisent, je vous envoie mes meilleures 73 ! 📻✨ (belle coïncidence, non ?)
13
6
68
975
Pierre-yves Baloche retweeted
Next Wednesday
13
23
6
201
10,762