Mastodawn

Por tanto desorden empece a probar #Picoclaw, la copia de #Openclaw pero china hecha con #golang.
Funciona mucho mas estable.

He subido un Pull Request integrando #Deltachat para comunicacion.

Si me pueden ayudar con su like en Github estaria increible:

https://github.com/sipeed/picoclaw/pull/2167

Emile `iMIl' Heitor Mar 23

For those of you with low-end GPUs (older 4GB NVidia card), nvidia/NVIDIA-Nemotron-3-Nano-4B-GGUF does a pretty good job at tooling, i.e. for use with smolclaw, @SipeedIO #picoclaw and friends

Kshitij Aggarwal Mar 23

Running two PicoClaw instances on one machine looked easy at first.

I thought duplicating the repo would be enough.
The real blocker was Docker runtime identity:
- container_name collisions
- shared bot token conflicts
- possible host port clashes later

What worked:
- separate repos
- separate data/config
- no container_name
- explicit Compose project names like picoclaw1 and picoclaw2

Short write-up here:
https://funkyidol.in/blog/deploying-multiple-picoclaw-instances-on-a-single-machine-with-docker/

#Docker #SelfHosting #DevOps #LocalAI #PicoClaw

Emile `iMIl' Heitor Mar 13

I ran pinchbench github.com/pinchbench/s... (manually as it has hard dependencies on #openclaw) on smolClaw/#picoclaw with local inference unsloth/Qwen3.5-35B-A3B-GGUF:UD-IQ3_S on an RTX5080. 95% success, here are the results: smolbsd.org/claw/pinchbe...

GitHub - pinchbench/skill: PinchBench is a benchmarking system for evaluating LLM models as OpenClaw coding agents. Made with 🦀 by the humans at https://kilo.ai

PinchBench is a benchmarking system for evaluating LLM models as OpenClaw coding agents. Made with 🦀 by the humans at https://kilo.ai - pinchbench/skill

GitHub

awlnx Mar 13

Documented my little adventure of running #picoclaw completely with a local #LLM.

https://gist.github.com/awlx/6387c4841ec1cef99621293ef272751a

Maybe it helps some people, I did it out of curiosity - how good a pure local setup with commodity HW would perform

And tbh it's not bad but bigger models would be better.

PicoClaw Private AI Assistant Setup

PicoClaw Private AI Assistant Setup. GitHub Gist: instantly share code, notes, and snippets.

Gist

Emile `iMIl' Heitor Mar 13

#picoclaw has now official NetBSD support! github.com/sipeed/picoc... is merged!

add NetBSD to the list of released platforms by iMilnb · Pull Request #434 · sipeed/picoclaw

📝 Description Add NetBSD to the list of released platforms 🗣️ Type of Change 🐞 Bug fix (non-breaking change which fixes an issue) [*] ✨ New feature (non-breaking change which adds functionality) ...

GitHub

Emile `iMIl' Heitor Mar 13

New smolClaw ships with uv. Why? because smol's going to pass the pinchbench tests! + all the fixes and features from #picoclaw ofc <3 github.com/NetBSDfr/smo...

smolBSD/service/clawd at main · NetBSDfr/smolBSD

smolBSD is a tiny BSD UNIX (NetBSD) system creation tool, primarily aimed at building modern, lightweight, fast micro VMs - NetBSDfr/smolBSD

GitHub

Emile `iMIl' Heitor Mar 12

A wild smolClaw landed! Latest innovations from #picoclaw integrated, including the long awaited web GUI! An isolated claw is a good claw github.com/NetBSDfr/smo...

smolBSD/service/clawd at main ...

smolBSD/service/clawd at main · NetBSDfr/smolBSD

smolBSD is a tiny BSD UNIX (NetBSD) system creation tool, primarily aimed at building modern, lightweight, fast micro VMs - NetBSDfr/smolBSD

GitHub

Emile `iMIl' Heitor Mar 10

smolClaw/#picoclaw running fully locally with a simple GTX 1650/4G: llama.cpp/CUDA Jackrong/Qwen3.5-2B-Claude-4.6-Opus-Reasoning-Distilled-GGUF:Q4_K_M is the biggest that fits Recipe gist.github.com/sudoingX/c2f... 70 tps!!! Build your own: github.com/NetBSDfr/smo... github.com/NetBSDfr/smo...

Patched Jinja template for Qwen 3.5 27B - fixes developer role crash + preserves thinking mode (thinking = 1). Drop-in replacement for agent tools (OpenCode, Claude Code, Continue, Cursor, Aider). Without this patch, --chat-template chatml silently kills thinking mode.

Patched Jinja template for Qwen 3.5 27B - fixes developer role crash + preserves thinking mode (thinking = 1). Drop-in replacement for agent tools (OpenCode, Claude Code, Continue, Cursor, Aider). ...

Gist

Emile `iMIl' Heitor Mar 8

uh oooh... github.com/sipeed/picoc... #picoclaw #IRC