Skip to main content

Documentation Index

Fetch the complete documentation index at: https://fromhazel.ai/docs/llms.txt

Use this file to discover all available pages before exploring further.

​
Your brand’s knowledge base

Think of Hazel like a brilliant new hire. Incredible general knowledge about ecommerce, marketing, and analytics — but doesn’t know your brand yet. Memory is the notebook that carries knowledge between conversations. When you teach Hazel something, it remembers. Every future conversation gets smarter because of what you’ve shared. Memory is one of Hazel’s most powerful features. Use it to store brand guidelines, goals, custom metric calculations, and anything else that makes your answers more accurate and tailored to your business.

​
Two types of memory

Org Memory

Shared across your whole team. When anyone adds something, everyone benefits. This is where your company’s shared knowledge lives.

User Memory

Personal to you. Only affects your conversations. This is where your individual preferences and role context live.

​
Org Memory

Org Memory is shared knowledge that helps Hazel understand your business, no matter who’s asking. Great for:
  • Metric definitions — what your team means by “active subscriber,” “returning customer,” or “blended ROAS”
  • Product categories — how you group your products, which ones are hero SKUs
  • Brand guidelines — your brand voice, seasonal calendar, important dates
  • Business rules — how you define customer segments, lifecycle stages, and thresholds
  • Goals and targets — revenue targets, CAC goals, retention benchmarks

​
User Memory

User Memory is just for you. It helps Hazel tailor its responses to your role and preferences. Great for:
  • Role context — your title, what metrics matter most to you
  • Report preferences — how you like data formatted, your favorite breakdowns
  • Communication style — whether you want bullet points, tables, or narrative summaries

​
What to teach Hazel

Here are some examples to get you started:
“When I say ‘active subscriber,’ I mean anyone with an active subscription who has placed an order in the last 90 days.”“Our blended ROAS (return on ad spend) includes Meta, Google, and TikTok ad spend. Don’t include organic.”“LTV (lifetime value) should be calculated as total revenue per customer divided by total unique customers for a given cohort.”
“Our hero product is the Original Tote. It comes in 15 colors and retails for $89.”“We have three product lines: Essentials (under 50),Core(50), Core (50-100),andPremium(100), and Premium (100+).”
“We consider a customer ‘lapsed’ if they haven’t ordered in 120+ days.”“Our fiscal year starts in February. When I say ‘this year,’ I mean starting Feb 1.”“We run two major sales per year: Summer Sale in June and Black Friday / Cyber Monday.”
“I’m the VP of Marketing. I care most about CAC (cost per acquisition), ROAS (return on ad spend), and email revenue.”“I manage our CX team. I’m focused on first response time, resolution rate, and CSAT scores.”

​
Teach Hazel about your data history

One of the most impactful uses of Memory is documenting historical events that affect how your data should be interpreted. Without this context, Hazel might misread patterns in your data.
If you migrated between platforms (e.g., from one subscription provider to another, or consolidated multiple Shopify stores into one), tell Hazel when the migration happened and how it affects the data.Example: “We migrated from Recharge to Skio in September 2025. Historical subscription data before that date is in Recharge. After September, all subscription data is in Skio.”
When importing historical orders into a new Shopify store, some date fields (like “Created At”) may get overwritten with the import date. Tell Hazel which date fields are reliable and which aren’t.Example: “We consolidated 3 Shopify stores into one in May 2025. The ‘Order Processed At’ date is accurate, but ‘Order Created At’ shows the import date for historical orders. Always use ‘Processed At’ for date-based analysis.”
Events like 3PL changes, payment processor issues, or significant spending shifts can distort trends. Documenting these helps Hazel contextualize anomalies.Example: “We changed 3PL providers in November 2024. Fulfillment quality was poor for 3 months before we switched, which drove up support tickets and refunds during that period.”
If you routinely exclude certain orders from reporting — wholesale accounts, internal test orders, free samples — add those rules to Memory so they apply automatically.Example: “Always exclude orders tagged ‘wholesale’ or ‘internal-test’ from DTC revenue calculations. Exclude gift cards and free samples from product performance analysis.”

​
How to manage memory

Click Memory in the side nav to add, edit, or remove entries anytime. You can switch between Org Memory and User Memory tabs to manage each type separately.
The more you teach Hazel, the better it gets. Most teams add 5-10 memory entries in their first week and keep building over time. You can also add memory right from a conversation — just tell Hazel “Remember that…” and it’ll save it.
Next up: Turn any question into an automated report with Scheduled Reports.