Ask HN: What is your current AI stack looking like?

Claude armed with Nano banana for all internal stuff. Deepseek for anything public facing.

Claude/GPT/Grok subscriptions, used together in Happy harness (I'm affiliated with Happy, but would probably use it regardless, it's free and opensource).

I'd also like to include local models in my workflow (for some simpler tasks), but currently just experimenting with that using LM Studio.

Claude Code + DeepSeek V4.1 flash + GLM 5.3

I use DeepSeek and GLM together for cross-validation. It keeps accuracy high and costs super low.

[flagged]