NewHolistics AI - now in beta

Ask your data.
Get answers you can trust, verify, and interact.

AI-assisted self-service analytics that lets anyone explore data with natural language. Grounded in governed metrics, verifiable logic, and your team's shared definitions.

Q3 Revenue Dashboard
7d30dQTD
Revenue
$2.84M
+12.4%
New users
18.2K
+8.1%
Churn
2.1%
−0.4%
Weekly revenueQTDthis quarterlast quarter
Revenue by region
$1M$500K$0
$380K
$560K
$700K
$920K
$240K
LATAM
EU
APAC
US
MEA
Ask AI
Why is revenue up this quarter?
Reasoning
Readendorsed metric weekly_revenue · 12 weeks
Filterperiod = QTD
Comparevs Q2 baseline
Decomposeby region & segment
QTD revenue is +12.4% vs Q2 (+$310K).
The US drove most of the lift
4 new enterprise deals closed in August (+18%).
New users (+8.1%)
Lower churn (−0.4%) compounded the gain.
grounded in weekly_revenueview AQL →
Ask a follow-up
Conversation

Query your data in plain English

Ask questions naturally and get instant answers. The AI keeps context across follow-up questions - so exploring data feels like a conversation, not a query builder.

01 - Conversational

Conversational data exploration

Ask questions in natural language. The AI remembers context from question to question, so you can drill deeper without starting over.

multi-turn memoryfollow-upsdrill-down
Ask AIConversation
Context preserved
M
What were our top 3 channels last month?
Paid Social $412K · Organic $338K · Email $214K
Metricchannel_revenueTimelast 30 days
M
Just paid social — break out by network.
$236K
Meta
$112K
TikTok
$64K
X
Metricchannel_revenueFilterPaid Social
M
Why is Meta up?
Ask AIDiagnostic· 3 hypotheses · 12s
INVESTIGATINGWhy did EU revenue drop last week?
RULED OUT
Seasonality
Same dip as Jul '25?
Jul '25 dip: −3%
MATCH
Product mix
Apparel category in DE?
−42%
apparel · DE
RULED OUT
Marketing
Meta spend −30%?
Meta spend: −7% w/w
ROOT CAUSE
Germany: best-selling apparel out of stock during promo
→ lost 4 sales days
02 - Diagnostic

Diagnostic analytics

Go beyond 'what happened' to 'why it happened.' The AI acts as your analyst - forming hypotheses, running queries, and surfacing root causes automatically.

hypothesisroot causeauto-investigate
The AI does far more than just fetch answers - it actually understands the context and generates working solutions, even for edge cases. It's saved me a ton of time already.
Guido Stark
Guido Stark
Head of Data, World Vision
Dual interface

Chat when it's natural.
Click when it's convenient.

A dual interface that meets you where you are. Start with a question in chat, then click into the generated chart to modify, filter, or explore further - no context lost.

Weekly revenueQTD
Apparel‑DE stockout · W7
Ask AI
Why is revenue dropping?
EU softness from W7 onward — Apparel‑DE stockouts account for −62% of the drop.
Two SKUs (DE‑J412, DE‑J418) hit zero inventory W6 Friday and haven't restocked. US & APAC are flat — this is regional, not global.
hover any chart point to reveal Ask AI
Ask AIthread #1
Why is revenue dropping?
EU softness from W7 onward — Apparel‑DE stockouts account for −62% of the drop.
Two SKUs (DE‑J412, DE‑J418) hit zero inventory W6 Friday and haven't restocked. US & APAC are flat — this is regional, not global.
EU
−38%
US
+4%
APAC
+7%
grounded inweekly_revenue
Ask a follow-up
Trust

AI you can actually verify

Every AI-generated answer is transparent. See the reasoning in plain English, then inspect the underlying query logic - so you never have to take the AI's word for it.

01 · Reasoning

Plain-language explanations

The AI restates its understanding of your question and walks through its reasoning step by step in plain language - so you can confirm it got you right before looking at the answer.

Ask AIReasoning
I understood your question as:
Compare EU revenue last week against a normal week, then find the product line driving the drop.
endorsed: weekly_revenue
Look at how EU revenue moved last week
Compare it to a typical week (median of the prior 4)
Break it down by product line
Surface the biggest contributor to the drop
Workspaceweekly_eu_change.aql
1
2
3
4
5
6
7
8
9
metric WeeklyRevenue =
sum(orders.amount)
by week(orders.created_at)
 
// Why: EU revenue -12% last week
EuWeeklyChange = WeeklyRevenue
| where orders.region = "EU"
| compare(percent_change, vs: median(prior 4 weeks))
| breakdown_by product_line
metric definition
composable pipe
02 · AQL

Readable query logic

See a readable, high-level query formula - not raw SQL - to verify exactly how every metric is calculated. Compact enough for a human to review in seconds.

03 · Lineage

Governed by your definitions

The AI prioritizes your team's endorsed metrics and composes new ones from existing definitions - not invented from scratch. Every answer traces back to your source of truth.

Ask AI"weekly revenue trend"
Endorsed only
From catalog · Endorsed
weekly_revenue
sum(orders.amount) by week
Built fromorders.aml · regions.aml
Used by12 dashboards
Not invented from scratch
ai_calc_revenue_v2
sum(orders.total) / 7
No lineage
Unverified definition
I'm seriously impressed with Holistics AI capabilities! While using Search Docs Mode, I realised it does far more than just fetch documentation - it actually understands the content and generates working solutions, even for edge cases not directly covered. Since then, I've been pasting in dashboard and theme code, and it consistently helps me fix or enhance things. It's saved me a ton of time already!
Guido Stark
Guido Stark
Head of Data at World Vision New Zealand
Built for AI reasoning

A query language built
for AI reasoning

Instead of translating natural language directly to SQL, the AI generates AQL - a high-level, composable analytics language. The result: answers that are more reliable, more readable, and more capable.

01Natural languageuser input
"In each region, what's the average revenue of the top 5 countries by users?"
AI · probabilistic
02AQL6 lines · human-readable
metric count_users = count(users.id)
dimensions { region: country.region }
measures {
avg_revenue:
top(5, countries.id, by: count_users) | avg(revenue)
}
Compiler · deterministic
03SQL55 lines · auto-generated
WITH "aql__t3" AS (
SELECT "ecommerce_countries"."continent_name"
AS "ecommerce_countries→continent_name",
"ecommerce_countries"."name" AS "name",
COUNT("ecommerce_users"."id") AS
"count_ecommerce_users→id"
FROM "ecommerce"."users" "ecommerce_users"
JOIN "ecommerce"."countries"
no SQL gymnastics
AQL
01

Verifiable & readable

AQL is compact and high-level, making it easy for humans to read and verify the AI's logic - no deciphering 200-line SQL queries.

02

Reliable

The AI works at a high-level language and doesn't worry about database-specific syntax or SQL gymnastics. The AQL compiler handles translation deterministically.

03

Composable & capable

Break complex operations into modular pieces and combine them like building blocks. Prebuilt functions handle period comparisons, running totals, nested aggregations, and more.

metric VolatilityIndex =
stddev(WoWChange) | over last 13 weeks
metric WoWChange =
WeeklyRevenue | compare(percent_change, vs: prior_week)
metric WeeklyRevenue =
sum(orders.amount) by week(orders.created_at)
metrics inside metrics · nest as deep as you need
I'd score AQL a 9 or 10. It's up there with the best tools I've used. With AMQL, it's really cool that we can now define metrics based on other metrics, stacking them on top of each other.
Ian
Ian
BI Engineer, DNSFilter
As code

AI works natively
with Holistics —
because all analytics
artifacts are code

Models as Code
Datasets as Code
Metrics as Code
Dashboards as Code

Models, datasets, metrics, and dashboards are all defined as text-based code. This means AI tools can read, generate, and modify your analytics definitions - and everything gets version control and governance for free.

AMLmodels.aml
AMLdatasets.aml
AMLmetrics.aml
AMLdashboards.aml
IDE

First-class IDE support

Use Cursor, Claude Code, or your preferred AI coding tool to develop models and dashboards faster than ever.

finance_dashboard.aml
weekly_revenue.aml
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
Dashboard finance_dashboard {
title: 'Financial Dashboard'AI ⌥↵
description: ''
block t1: TextBlock {
content: @md <div
style="
width: 100%;
height: 100%;
background: linear-gradient(to
right, black, #021
background-size: cover;
background-position: top-
center;
background-repeat: no-repeat;
padding: 12px;
border: 2px solid #EDF1F5;
border-radius: 8px;
">
</div>;;
theme {
background {
bg_image: 'linear-
gradient(97deg, #051527 20%'
}
Git

Version control & governance

Every change is tracked. Review diffs, manage branches, and enforce governance workflows - the same way engineering teams manage code.

Pull request · #247
main
feature/ai-attribution
+ metric AttributedRevenue = …
− metric Revenue_v1 = sum(…)
Docs

Self-documenting semantic layer

AI writes field descriptions, dataset summaries, and tag descriptions - so your semantic layer stays documented without manual overhead.

Without AI
orders
users
products
revenue
With AI
ordersApproved
usersMarketing
productsSales
revenueFinance
Workflow

Auto-generated commits & PRs

Auto-generate commit messages and PR descriptions. Focus on the analytics work, not the paperwork.

Commit changes (1)
Create PR
Commit message *AI
Update title of 'Financial Dashboard' to 'Executive Dashboard' in finance_dashboard page
finance_dashboard.page.aml
Commit
I really loved that you have all of that in code. When working with ChatGPT, you can just say 'here's the table I have, can you add descriptions to everything?' It makes AI actually useful because it can parse and understand your models.
Tina Einsiedler
Tina Einsiedler
Tandem Health
AI
Semantic models
Org instructions
Per-user context
Context

Teach AI your business - precisely

AI starts smart by drawing from your existing models and definitions. Then go further: customize its behavior at the organization, team, or individual level with programmable context.

01

Rich context by default

AI automatically draws from your semantic models, metric definitions, dashboards, and conversation history. No manual setup required to get accurate answers from day one.

02

Organization-level instructions

Define your business background, operational knowledge, and response preferences in a central config - so AI speaks your company's language and follows your standards.

03

Dynamic per-user context

Customize AI behavior per team or user with programmable logic. Marketing sees marketing datasets, purchasing sees purchasing datasets - automatically.

The dynamic context - where you can customize AI behavior per user or team - that's really cool. Other tools can't do that easily.
Jerry B.
Jerry B.
Cortex
Distribute

Bring AI analytics everywhere

Connect to your existing tools or embed AI-powered analytics directly into your product - no AI stack required on your end.

MCP Server

Holistics MCP Server

Use Holistics' querying power inside any external AI tool via the Model Context Protocol. Your AI assistants can query governed metrics natively.

HolisticsMCP
Cursor
Claude
ChatGPT
Custom
governed metrics, queried natively
Embedded

Embedded AI Chat

Give your customers AI-powered analytics inside your product. No need to build your own AI - embed the Ask AI experience directly.

acme-shop.app/insights
Acme Shop
Revenue
$84.2K
▲ +12%
Orders
1,247
▲ +8%
AOV
$67.50
▲ +4%
Daily revenue · last 14 days
AskAcme AI×
Which SKUs are hot this week?
Indigo Wallet+180%
Linen Tote+94%
Espresso Cup+62%
Ask anything…

Frequently asked questions

How to activate Holistics AI?+

To join our beta program for Holistics AI, please visit the following link: Join our waitlist. We will notify you once the beta is ready.

How does Holistics AI use my data?+

To provide Holistics AI functionality, we share your modeling metadata with third-party AI providers while you use the AI features. Rest assured, these providers or contractors are not permitted to use your data for training their models, and we do not utilize the AI inputs and results from our customers to enhance our service.

What models does Holistics AI leverage?+

Currently, we use Open AI API, adhering to their policy of not utilizing prompts or any inputs as training data. Additionally, we are experimenting with other models to identify those that offer the best performance and user experience.