Local AI setup on an RTX 4080: split the jobs across models, keep notes in Markdown
A user with an RTX 4080 SUPER and 64GB of RAM asks which local models and harnesses fit deep research, quick lookups, notes and task tracking. The useful answer: split the jobs across models, keep notes in plain Markdown the agent can re-read, and give calculations to a real calculator tool instead of letting the model do maths in its head.