Been running a fully local web research pipeline for my AI agent setup for a while now and realized I haven't seen much discussion about how others are handling this part. The inference side gets all the attention, but getting an agent to actually browse the real web without everything falling apart
