Annabelle doesn't guess. She reads from a growing library of pre-verified market research, integration docs, and platform knowledge — loaded fresh at the start of every session.
API status, auth models, rate limits, and partnership requirements for every platform we've researched. ServiceTitan, QuickBooks, BuilderTrend, and more.
Pool industry context: who the players are, what software they run, and where Stinger fits — so Annabelle can brief your team without generic AI hallucination.
What Annabelle can and can't do, kept current. She knows her own abilities because they're documented here — not because she's guessing from training data.
Not summarized from AI output. Not inferred from documentation summaries someone else wrote. Every entry in the vault was checked against the authoritative source — the vendor's developer docs, the actual API response, the real rate limit header.
Every entry links to its source. The date of verification is recorded. When something changes — a vendor deprecates an endpoint, a rate limit drops — the entry gets updated and re-dated, so you always know exactly how fresh the information is.
Every agent, every conversation, reads the same ground truth. Annabelle's sub-agents, the Teams room, and Claude sessions all draw from the same vault index — so there's no drift between what one agent thinks a platform can do and what another one says.
Most AI assistants hallucinate integration docs. They describe APIs that were deprecated two versions ago, suggest OAuth flows that don't match the current SDK, and fabricate partner program details. The vault exists to prevent that. When Annabelle says "ServiceTitan has a 60 req/sec rate limit," it's because we verified it from their developer documentation — not because a language model estimated it.