Two AI stacks that do different jobs

This past week I’ve been tinkering with 2 AI stacks.

The first is Hermes + GLM, running on Telegram. Hermes is an agent you can ask to do all kinds of things: handle client replies, draft Threads content, do research, even automate workflows that are usually manual or repetitive.

I’m learning it slowly, one use case at a time. What it costs me is a VPS and tokens, and it’s still worth it.

The second is Opencode + DeepSeek + Obsidian. It’s for tidying up my personal notes, project management and a lifewiki that’s been a mess all this time.

I only topped up $2 of DeepSeek credit, and I’m still using it now.

Yesterday I talked it through with my mentor, and the advice wasn’t to add anything. Quite the opposite: audit it all again and declutter whatever doesn’t connect to my goals, so everything lines up better with where I’m heading.

That’s what I learned this week: these two stacks do different things.

Hermes + GLM add capability. Opencode + DeepSeek reduce load. We often focus only on the first, when the second matters just as much.