The vault is free forever.
The chat is at-cost.

A closed corpus — twenty-five of Lacan's Seminars and fifty volumes of secondary literature, and growing — indexed across eleven thousand passages, queried with citations back to the page. Every answer comes from the actual text. No hallucinations. No filler.

The vault runs deep. Each of the 216 canonical concept pages averages roughly 50,000 words — long-form analysis per concept, not summary, organised across Definition, ELI5, Place-in-the-corpus, Key Formulations, Cited Examples, and Tensions sections. The lighter novel concept pages cover several thousand more concepts that surface from the texts. The chat queries that depth on every turn — what you're paying for is access to the analysis, not a stripped-down summary.

Each question hands the model a large window of retrieved passages plus your conversation history — typically 30–60K tokens of context per turn, sometimes much more on deeper threads. That's how cross-corpus answers stay grounded. It's also why each turn costs what it costs: feeding the latest Opus models that much context is expensive to run.

Less context would be cheaper and shallower. Rationing what the model sees would make the answers feel more like a generalist chatbot — short, confident, partly wrong. Your ledger pays for the full context window instead. When it empties, the session ends. No ad model, no data resale, no margin extraction baked into this — just the actual cost of running the chat, plus enough buffer to keep it online.