llmwiki compiles papers into persistent wiki

The pitch is strong because it attacks the real failure mode of most AI research workflows: not answering once, but preserving the synthesis so the next session starts ahead. This looks more useful than another chat layer, but only if users are willing to maintain the wiki as a living artifact.

• –The `compile`, `query --save`, `watch`, and `lint` loop is the right shape for compounding understanding over time.
• –Shipping an MCP server makes it easier to plug into agent workflows, which is where this idea gets materially better than plain document chat.
• –The upside is best for literature reviews, notes, and team memory where structure and cross-links matter more than raw retrieval.
• –The main risk is stale or noisy knowledge if the compiled wiki is not reviewed and curated; persistent memory is only an advantage when it stays trustworthy.
• –This is squarely in open-source AI tooling, not a model release, and it will live or die on workflow fit rather than novelty.