codinglog

Hit blocks doing research from the VPS-hosted Hermes agent — web search and extraction kept getting blocked. The unblocking stack: SearXNG for self-hosted meta-search, Jina AI for extracting pages as markdown, and Camofox as a stealth browser with VNC attached for visual inspection. The goal is simple: enough tooling for a smooth research-supported assistant without getting IP-blocked or bot-detected on every call. Each piece solves a specific failure mode — SearXNG aggregates search engines (no API keys, no shared rate limits), Jina handles the "give me this page as readable text" job, and Camofox handles the "site refuses non-browser clients" job.

Having that feeling that agent upkeep is a hidden layer of work nobody accounts for when setting these up on a VPS. It's easy to say "just deploy it" — the complexity that comes after is the real cost. Every tool bolted on to unblock a problem becomes its own thing to maintain. SearXNG, Jina, Camofox, VNC, Directus, the VPS itself — suddenly the agent isn't an agent, it's a small platform with a platform's maintenance load. The honest question: is all this worth it for the benefits, apart from the learning (which is genuinely good)? The trap is that infra work feels productive but doesn't produce content. If upkeep starts eating the time it was supposed to free up, that's the signal to simplify.

Hermes is live on Oracle Cloud. Created profiles for my brother. Now using DeepSeek V4 as default model via Neural Watt — will monitor usage and results, then decide if we need to switch to another model or provider.

Spent the last day getting LiteBin to work with remote agent node deployment. Deploying Hermes on Docker helped me see the gaps in LiteBin's agent node deployment — now I'm fixing them.