Analytics Assistant

How it works

One context layer. Three ways to ask.

The context layer is built from your Semantic Nexus

We sync your dbt models, Cube definitions, metric glossary and business rules into the assistant's context: which table is the source of truth, how revenue is defined, which time column to filter on, which clients are archived. The assistant knows what your best analyst knows — and a weekly sync keeps that knowledge current.

Anyone asks, wherever they already work

Mention the assistant in a Slack channel, open the web chat, or add it to Claude as a connector with a single URL. Threads keep their memory, so a follow-up like “now by store” or “top five products in June” just works. No SQL, no ticket, no queue.

Every answer is accurate, explained and shareable

Each answer comes with a chart, the CSV behind it, and a short note on scope and methodology. The generated SQL is one click away for anyone who wants to check the logic. A thumbs-down goes straight to the context owner, so the context improves question by question.

Example output

A question asked in Slack, answered in the thread.

Illustrative example using a fictional coffee-shop chain. The assistant answers with a chart, the data behind it, and how it got there.

#analytics

Maya K. Customer Success · 9:12 AM

@assistant what was total revenue in June, broken down by store?

Analytics Assistant APP 9:12 AM

Total revenue in June was €175.3K across three stores. Hell's Kitchen led at €62.1K; Astoria and Lower Manhattan were within €2K of each other.

Revenue · June · by store EUR

€57.4K

€62.1K

€55.8K

Lower Manhattan Hell's Kitchen Astoria

🧠 Scope & metric · Gross allocated turnover in EUR, 1–30 June, no-sales excluded, active stores only.

📐 Methodology · Sum of order-line turnover per store, rounded in SQL. Query available in nao.

revenue_by_store_june.csv Open in nao 👍 2 👎

Follow-up in the same thread

Maya K. 9:14 AM

and the top 5 products that month, with units sold?

Analytics Assistant APP 9:14 AM

Top 5 products in June, all stores. Latte alone is 18% of revenue.

product revenue units

Latte€31.6K6,740

Cappuccino€24.9K5,410

Chai latte€17.2K3,580

Cold brew€14.8K3,290

Espresso€11.3K4,120

What you get

Everyone data-driven. One source of truth.

Governed context layer

Metric definitions, business rules and table knowledge from your dbt and Cube models, synced weekly. Two people asking the same question get the same number — and a new colleague is onboarded the moment they ask.

Slack-native answers

Mention the assistant in any channel and get the answer in the thread, with a chart image and CSV, visible to everyone. Threads remember context. Scheduled posts deliver recurring numbers, like Monday's gross turnover, without anyone asking.

Claude connector

Add the assistant to Claude as an MCP connector by pasting one URL. Brainstorm with your data in a full conversation — “which product grew most between Q1 and Q2?” — and get the same numbers as in Slack, because it's the same context layer.

Tested accuracy

Your key metrics are covered by tests that run plain-language questions through the assistant and compare the result with the expected SQL. Every answer exposes its query, and a thumbs-down routes to the context owner for review.

Stories: shared live dashboards

Turn a conversation into a dashboard. Reorder or drop charts, add commentary, and share it with the team or keep it private. The story stays live against your data, so it's a page that updates rather than a screenshot that ages.

Your infrastructure, your model

Self-hosted on Google Cloud Run in your region, with invite-only access and role-based shielding of sensitive slices. Built on nao, an open-source framework, with any LLM behind it via your own API key. No per-seat licence, no vendor lock-in.

Who it's for

The right fit.

Companies that have a Semantic Nexus, or are building one, and far more people with data questions than people who can write SQL. Sales and customer success preparing for a client conversation. Leadership working through a strategic question and wanting the number in twenty seconds, not next week. Support and operations checking what the client said against what the system actually recorded.

It works best when the metric definitions are already agreed. The assistant makes them accessible to everyone; it doesn't invent them. That's exactly what the Semantic Nexus is for.

Roadmap

What's coming next.

Next up

Direct Cube connection

Today the assistant reads your Cube definitions into its context and queries the warehouse. Next it plugs straight into the Cube semantic layer, so measures, dimensions and access rules live in exactly one place and the assistant inherits them automatically.

Coming later

Specialised agents on one context

From a general-purpose analyst to a set of focused assistants for finance, customer success and operations, each with its own skills and shielding, all sharing the same governed context layer. The governance stays central; the expertise gets specific.

Pricing

Implementation plus a flat licence.

Two components. An implementation fee covers building the context layer from your Semantic Nexus, writing the accuracy tests for your key metrics, and wiring up Slack and Claude. A monthly licence covers hosting, model usage, and keeping the context maintained and in sync.

There is no per-seat pricing. The assistant is meant to be used by everyone in the company, so the licence is flat rather than per user. We scope the exact structure together based on your data model and the number of metrics under test.

Implementation fee

Context layer, metric tests, Slack and Claude setup.

Monthly licence

Hosting, model usage, context maintenance. Flat, not per seat.

Get a scope

Built and maintained by

The people behind the Analytics Assistant.

Sarah Thiébault

Product Owner · Maintainer

Medior Data Engineer · Maxq Analytics

Sarah leads the Analytics Assistant end to end. She built and deployed the first version, runs the context layer and its accuracy tests, and owns the roadmap and client rollouts. She is the main point of contact for the add-on after implementation.

Philip Boontje

Product direction

Guild Master · Maxq Analytics

Philip shapes how the assistant answers — short, consistent, chart-first — and connects it to the Semantic Nexus architecture it runs on, so the context layer is a by-product of the data model rather than a second thing to maintain.