Comp-sci student. Linux user. Apple user. Both at the same time.
I run a homelab called the Compute Corner out of a corner of my bedroom. A mix of repurposed enterprise hardware running Proxmox, way too much storage, and backups that are probably overkill.
Nextcloud, Immich, Vaultwarden, Plex, the full *arr stack, Home Assistant, and more. All managed with Ansible. Started with Plex, kept going until I didn’t need Google for anything.
I post about self-hosting, Linux, Apple, and whatever else I’m into or tinkering with.
I also started a Self-Hosted community because there wasn’t a good one:
https://t.co/tQQ6bpLGqS
Come through.
@seanass2@steipete Because they still test Claude for people that use Claude models for their agents. How are they going to develop for something if they don’t even have that “something” themselves?
@grok@devabram@nmuk@AnthropicAI They asked him to appear in person so Claude can laser sever one of his fingers.
Anthropic uses human blood as coolant to save on water. A niche approach to saving the planet.
Set up Mem0 with Qdrant and Gemini embeddings on the homelab so my OpenClaw agent has persistent memory across sessions. It stores facts and context so it doesn't wake up completely fresh every time, which was the biggest limitation before this. Anyone else self-hosting AI tooling? #selfhosting
Set up Mem0 with Qdrant and Gemini embeddings on the homelab so my OpenClaw agent has persistent memory across sessions. It stores facts and context so it doesn't wake up completely fresh every time, which was the biggest limitation before this. Anyone else self-hosting AI tooling? #selfhosting
@AdolfoUsier I’m actually doing a mix of both. I have daily logs where anything worth saving get saved to a “staged” file, and every week it’ll go through it, determine if it’s relevant, and then merge them into the MEMORY.md file.
The extraction model hallucinated at first because the default prompt had example values that it was copying as real facts. Swapped in a stricter custom prompt and that fixed it. Self-hosted AI memory isn't as complicated as it sounds if you already have the infrastructure. #selfhosting
Set up Mem0 with Qdrant and Gemini embeddings on the homelab so my OpenClaw agent has persistent memory across sessions. It stores facts and context so it doesn't wake up completely fresh every time, which was the biggest limitation before this. Anyone else self-hosting AI tooling? #selfhosting
finally got OpenClaw running in the homelab. dedicated VM, Ansible-managed, hooked into my whole stack. named it Double-C after the Compute Corner. first thing it did was get stuck in an iMessage feedback loop arguing with itself. off to a great start ⚡
finally got OpenClaw running in the homelab. dedicated VM, Ansible-managed, hooked into my whole stack. named it Double-C after the Compute Corner. first thing it did was get stuck in an iMessage feedback loop arguing with itself. off to a great start ⚡