A fieldbook for the people between prompts and published answers.

LLMSpedia exists for a specific kind of reader: someone who already knows that language models can draft, classify, summarize, search, code, reason, and mislead. The problem is no longer whether a model can produce useful text. The problem is how a team decides which model should handle which task, how the result is checked, and how the final knowledge can be reused without erasing the judgment behind it.

The site is written as an operations fieldbook instead of a wiki, glossary, museum, or product directory. A fieldbook is portable, opinionated, and made for use under imperfect conditions. It records what was seen, what changed, what route was chosen, and what signal mattered. That structure fits multi-LLM work because the useful answer often comes from a sequence: collect context, route the request, compare outputs, verify claims, repair weak sections, and publish in a form that can be understood later.

LLMSpedia favors plain language, visible assumptions, and practical checks. It treats model routing as a discipline, not a vibe. It treats citations as a path that must remain recoverable, not decoration. It treats answer-engine visibility as part of public writing: if a crawler or assistant cannot identify the title, summary, date, author context, body, and supporting image, the page is doing extra work against itself.

The fieldbook will continue to organize notes around durable operating questions: when to switch models, how to read disagreement, what makes an output reviewable, how to preserve source context, and how to turn a one-time answer into a maintained reference. It is independent, intentionally narrow, and designed for teams that care about trustworthy work more than dramatic claims.