Skip to content

Analytics agents need the context your team keeps in their heads.

Cassis is where it lives: a living context layer between your data and your business. Sharpened by every conversation, governed by your data team, fueling all your agents.

The problem

The context was always scattered and stale. Analytics agents turn that into production risk.

Every table, dashboard, and metric hides human decisions. Without that context, agents guess.

Business evolves, the data stack changes, new domains open. Without maintenance, agents drift.

# data-questions
Sophie P. 10:23 AM

Hey @data-team the board deck says 4,215 active customers but the revenue dashboard says 3,892. Which one do I use? Exec meeting in 2h

Marco R. 10:41 AM

Board deck pulls from Looker, anyone with a login in the last 90 days. Revenue dash uses Metabase, filters on paid_plan = true Both are "active customers."

😩 3💀 2
DataBot app 10:42 AM

Based on the customers table, you currently have 5,104 active customers.

🤦 4
Sophie P. 10:47 AM

Great, now we have four numbers. Which one goes on the slide.

The product

Cassis is where agents find the context your team trusts.

Before SQL gets written, Cassis turns the question into a trusted data path: the concepts involved, the definitions your team approved, the joins that are valid, and the rules that shape the answer.

Talk to us
How many active customers grew revenue last quarter?
Resolved to
Business meaning
active customer paid plan, status = active, excludes trials
grew revenue ARR Q over Q delta > 0
last quarter 2026 Q1, Jan 1 to Mar 31
Data path
metric customer_revenue_qoq_growth
sources fact_revenue_monthly, dim_customer
grain monthly to quarterly
filters status = active AND subscription_tier = paid
Produces
2,335 customers grew revenue last quarter
+$1.4M ARR growth, 60% of active base
How it works

Cassis builds and keeps your context alive.

Cassis bootstraps from your data stack and documentation. From there, every question, correction, schema change, and ambiguity enriches the ontology.

Day 1
Built from your stack.
Sources
dbt project184 models
Looker42 views
Snowflake8 datasets
Notion12 docs
Slack archive240 threads
Building ontology
Ontology review
87% auto-mapped 13% needs your call
56 entities · 142 metrics · 14 dimensions
Conflicts 12Missing 4Ambiguous 3
Every day
Continuously enriched.
Signals
conversation
4,215 active customers.
based on active status
That counts free trials. We only count paying customers.
active customer
status = 'active' AND plan =  'paid'
dbt plan renamed to plan_type
For data teams

Governance without becoming the bottleneck.

  • The queue handles itself.

    Related issues are grouped, traced to the definition or model behind them, and ranked by potential impact.

  • Drift surfaces before it breaks.

    Schema changes, rule changes, and rename collisions arrive as flagged updates, not Slack alerts from finance.

  • Every ontology object has receipts.

    Owner, source, edit history, last review. Roll back any change. Trace any answer to the context that produced it.

Review queue
3 open
New context acquisition_channel Proposed from 9 questions. Source: UTM fields and campaigns table.
Review
Clarify loyal_customer Two attributed definitions. Sales: 3+ renewals. CS: >12mo subscription.
Review
Schema drift invoices.discount_amount Column renamed to net_discount. Downstream definitions flagged before refresh.
Review
For everyone else

Numbers you can defend in the meeting.

Every answer shows its work.

What was revenue in Q4?
$2.45M
Definition Revenue, Finance
Excludes refunds + chargebacks
Reviewed Mar 11, S. Parker

Ambiguity surfaces. It does not hide.

What is our churn rate?
Two definitions exist. Which one do you mean?
Logo churn % of customers lost
Revenue churn % of ARR lost

Cross-domain questions get the joins right.

Which Q1 prospects converted to paid accounts by Q4?
127 prospects · $1.4M ARR
Marketing prospects
Finance accounts
Sales opportunities
Where it fits

Augments your trusted data stack.

Data platforms
WarehousesLakehouses
Modeling
dbtSemantic layers
Documentation
CatalogsDocs
Cassis agent
Conversational interface
AI clients
Claude CodeCodexDust
Agents and workflows
SlackAnalytics agents
Blog

Notes on data, meaning, and context

Become a design partner
Context engineering for analytics agents: lessons from six months of building and rebuilding
Context engineering

Context engineering for analytics agents: lessons from six months of building and rebuilding

Analytics agents need context. Great. How should you structure it? Lessons from six months of building, testing, and rebuilding it.

The state of agentic analytics, from 50 real data teams
Field notes

The state of agentic analytics, from 50 real data teams

Field notes from 50+ conversations with data teams: the five stages of agentic analytics, what breaks at each, and what teams want next.

The hard part of self-maintaining context for analytics agents
Context engineering

The hard part of self-maintaining context for analytics agents

Context is now table stakes for analytics agents. The new problem is keeping it true as the business keeps moving.

Start trusting youranalytics agents

Cassis is in early access. We're working with data teams at companies with well-structured data stacks (between 200 and 1000 people) who are ready to move beyond duct-taped context.