Customer Knowledge Vault
The factual grounding that prevents AI from making things up about your business — products, FAQ, target customer, banned claims.
What it is
The knowledge vault is structured per-customer business data — products, FAQ, target customer description, brand history, sendable assets, approved/banned claims — that gets threaded into every AI call alongside the voice profile. Without it, AI invents details about your business; with it, every draft is grounded in real facts.
Operators tired of AI tools hallucinating fake product names, made-up pricing, or invented customer stories; brands with specific compliance constraints (no income claims, no medical claims) that need enforcement; agencies managing content for clients with detailed brand bibles that have to be honored.
Why it matters
The specific outcomes this capability is designed to produce — and the failure modes it's designed to prevent.
Stops AI from inventing facts about your business
AI without grounding will invent product names, fabricate pricing, and reference customers that don't exist. The knowledge vault gives Claude the real catalog so it stops making things up. No more "we offer 3 service tiers starting at $499" when you actually have one tier at $97.
Sendable assets registry — drafter pulls real URLs
Add your products as labeled URLs ("Audit funnel — for service businesses considering a website refresh | https://your.site/audit"). Recon's drafter picks one per draft based on conversation context. No more [link] placeholders in drafts you have to fill in manually.
Compliance-aware claim enforcement
Add 'no income claims' / 'no medical claims' / 'no guaranteed results' to the banned-claims list. The drafter prompt enforces them — generated content that violates gets rewritten before you see it. Critical for coaches, financial advisors, MLM reps, anyone selling outcomes you can't legally guarantee.
Customer-specific FAQ in every reply
Riley's auto-draft replies pull from your FAQ. When a prospect DMs 'how much does it cost,' the auto-draft references your actual price, not a generic answer. Same for product questions, refund policies, integration questions.
How it works
From "I want to use this" to "it's running in production for me," in 4 steps.
Open the Knowledge Vault
At /dashboard/knowledge. Categorized form: products, FAQ, target customer, brand history, sendable assets, approved claims, banned claims.
Start with sendable assets + FAQ
5-10 sendable assets and 10-20 FAQ entries get you 80% of the value. Most operators fill these out in 30 minutes the first session.
Add banned claims for compliance
For coaches: ban "guaranteed results," "doubled my income," "you'll definitely." For health/wellness: ban "cure," "treat," "diagnose." The drafter rewrites violations.
Test by generating content
Generate a calendar. Check that referenced products are real, claims match your actual brand, and the AI isn't inventing details. Iterate the vault when you spot drift.
Questions
How is this different from the voice profile?+
Voice profile = HOW you sound (tone, phrases). Knowledge vault = WHAT you know (facts, products, claims). Both threaded into AI calls. Voice profile alone fixes "sounds generic." Knowledge vault alone fixes "AI made up a fake product." You want both filled in.
How big can the vault get before AI calls slow down?+
Practical limit: ~10K characters of vault context per call (about 3-4 pages of content). At that size, AI calls take ~3-6 seconds (vs ~2-3 without). Above 10K, we summarize older entries to keep tokens manageable. Most customers stay well under this — the highest-value content is FAQ + sendable assets, which together are usually 2-4K characters.
Can I import from an existing brand bible / Notion doc?+
Today: copy-paste manually. CSV import is on the roadmap. The vault categories map cleanly to most brand-bible structures so the copy-paste is usually 30-45 minutes for a comprehensive one.
Are banned claims actually enforced or just suggested?+
Enforced at the prompt level. The drafter prompt explicitly forbids each banned phrase. If the AI tries to use one, the prompt instruction catches it and produces compliant content. We've tested this against MLM income claims, coaching guarantees, and medical-language bans — all reliably blocked.
Can I see what the AI is reading from my vault?+
Today: not directly visible (the vault is threaded into Claude's system prompt, not shown to the user). If you want to debug a generated draft, the simplest path is to clear the vault temporarily, regenerate, and compare. Direct visibility is on the roadmap as a "preview AI prompt" feature.
You might also need
Capabilities that pair naturally with customer knowledge vault — each included with every Iron Front Digital subscription.
Brand Voice Profile
Train the AI on your tone, your phrases, your story — once. Every draft from then on sounds like you, not like AI.
Riley — AI Mission Control
The front-desk AI that triages, routes to specialists, and drafts on demand. Honest about being AI. Never pretends to be a real person.
AI Social Media Calendar Generator
Generate a 7-day Facebook + LinkedIn calendar with hook-led posts, voice-grounded drafts, and a built-in 7-day rotation that prevents pitch-fatigue.
How Iron Front compares
Already paying for ClickFunnels, HubSpot, GoHighLevel, or Vendasta? These pages walk through what overlaps, what doesn't, and where each tool genuinely wins.
Ready to use customer knowledge vault?
Included with every Iron Front Digital subscription. The faster you start, the more time the system gives you back.