Fredoline Eruo

Fredoline Eruo is the operator behind runlocalai.co — a structured catalog of local AI hardware verdicts, decision tools, and a growing community-benchmark feed. The site is independently operated; one person makes every editorial decision and signs every byline. The catalog focuses on practical questions: what models actually run on consumer hardware, how VRAM and quantization affect usability, and what real cost of ownership looks like compared to cloud equivalents. Recommendations are grounded in published methodology, computable math from spec sheets, and a small (but growing) set of measured benchmarks plus named community submissions. **Focus areas** - VRAM efficiency and quantization tradeoffs (GGUF, AWQ, EXL2) - Cost-of-ownership math (cost-vs-cloud, compounder TCO) - Local inference stack decisions (llama.cpp, Ollama, vLLM, MLX) - Interactive decision tools (will-it-run, stack builder, quant advisor) **Hardware bench — with published benchmarks** - RTX 3080 Laptop 16GB — Qwen 2.5 Coder 7B Q4_K_M at ~70 tok/s (community-benchmark submissions on /community) **Hardware owned — benchmarks queued** Additional devices are queued for the editorial benchmark backlog. Until they have measured tok/s on /community, they aren't cited as "tested." A site that calls out other operators' fabrication has to hold itself to the same standard. **Editorial principle** Every number on this site must trace to one of: (1) a DB-row spec we can audit, (2) computable math from those specs, (3) a cited community measurement, or (4) our own bench run with reproduction notes. Anything else doesn't get published — and when older entries fall short, they get hedged or pulled.