The narrative around local AI has an unfortunate hardware elitism problem. Articles about running models locally tend to assume you have an RTX 4090, a MacBook Pro M3, or at minimum 32GB of RAM. The practical reality is more accessible than this — meaningful local AI is achievable on machines that cost $400-600.
What Actually Works on 8GB RAM
Ollama with a quantised 7B model runs adequately on 8GB RAM. "Adequately" means responses take 5-15 seconds on CPU-only hardware — slower than cloud, but fast enough for non-real-time tasks like drafting emails, summarising documents, and answering questions. Models to try: Llama 3.2 3B (fast, fits in 4GB), Mistral 7B Q4 (good quality, fits in 5-6GB), Phi-3 Mini (surprisingly capable, very small). For a laptop with an integrated GPU (Intel Iris, AMD Radeon 780M), performance improves noticeably — these can accelerate model inference meaningfully even without a discrete GPU.
What 16GB Unlocks
With 16GB RAM you can run 13B models at good quality, or 7B models very quickly. This is where local AI starts feeling genuinely fast — 1-3 second responses rather than 10-15 second waits. A mid-range laptop (2022 or newer) with 16GB and any discrete GPU is a capable local AI machine. The sweet spot in 2026 for local AI capability-per-dollar is a used ThinkPad X1 Carbon or MacBook Air M2 — both run 13B models well.
Skales works on all of these configurations. Connect Ollama in Settings, choose your model, and every Skales feature works with that local model — email, calendar, file management, automation, and more. Read about fully offline operation.