Mobile Chat Setup with NVIDIA DGX Spark and Gwen3.6
Follow-up to the GX10 vLLM setup: running Open WebUI and SearXNG on a Synology NAS, accessing the local Qwen model from iOS over Tailscale, and keeping the raw model endpoint off the internet.
walterra's small website
Follow-up to the GX10 vLLM setup: running Open WebUI and SearXNG on a Synology NAS, accessing the local Qwen model from iOS over Tailscale, and keeping the raw model endpoint off the internet.
Deploying the aeon-vllm-ultimate container (Qwen3.6-27B + DFlash) on an ASUS Ascent GX10, learning the hard way how unified memory locks you out, wiring it into pi, and having the local model one-shot a Nokia-style Snake clone.
Notes from setting up an ASUS Ascent GX10 on a local network, getting ds4/DwarfStar running with DeepSeek V4 Flash, and wiring it into pi-coding-agent with my bot-prompt system prompt.
Notes on two Eddo releases covering persisted Telegram assistant history, pi-ai provider support, structured scheduling, and timezone fixes.
Conversational AI is anthropomorphic theatre. After too many 'You're right to push back' moments, I snapped and wrote a system prompt that turns AI agents back into machines. Here's why and how.
How I set up Qwen3.6-35B-A3B as a local coding agent on Apple Silicon using pi-coding-agent, with MLX server tuning for best performance.
Stable release of my Bluesky firehose to Elasticsearch CLI, with better docs, config, and examples.
Setting up Elasticsearch 9.3 + Kibana on a Synology DS925+ by running Docker inside an Ubuntu VM to get seccomp support.
A small update to my Elasticsearch data ingestion library with dual-version support and TypeScript definitions.
Two small CLI tools to find Mastodon/Fediverse accounts from the people you follow on X and Bluesky, then generate import-ready CSVs.