OpenClaw exposes CPU-only latency wall

This looks less like a weird edge case and more like the current ceiling of CPU-only agent workflows hitting hard. Local-first agents feel magical right up until they have to prefill thousands of tokens before doing any useful work.

• –The reported ~6,060-token system prompt is a huge prefill tax, and on CPU inference that cost often dominates time-to-first-token more than generation speed does.
• –Ollama has made OpenClaw dramatically easier to install with `ollama launch openclaw`, but setup simplicity does not remove the raw compute burden of 14B-32B models and 8k-16k contexts.
• –The thread highlights a product gap around persistent KV caching, slimmer default instructions, and CPU-tuned model presets for agentic use cases.
• –For developers building local AI stacks, this is a reminder that privacy and control are real wins, but sub-minute agent loops usually still need much tighter prompts, smaller models, or GPU help.