How Compass Works

How Compass Understands Your Data

Compass is designed to work directly with your data warehouse and learn your business context through natural conversation. Here's how:

1. Direct Connection to Your Data

When you connect a data source, Compass immediately begins reading your database metadata—table structures, column names, descriptions, and relationships. This means you can start asking questions right away, without any upfront configuration.

What makes this different:

• No semantic layer required - Unlike traditional BI tools, you don't need dbt models, LookML definitions, or cube configurations to get started
• Zero setup time - Connect your warehouse and start exploring immediately

See Data Sources for supported warehouses and connection instructions.

Note: If you already use a semantic layer, we're building integrations to leverage your existing definitions. See Semantic Layer Integration below for details.

2. Auto-Generated Documentation

Compass automatically creates documentation for your tables and columns based on your database metadata. When you connect a warehouse, Compass captures:

• Table and column names - Understanding the structure of your data
• Data types and constraints - Knowing what kind of information each field contains
• Existing descriptions and comments - Leveraging any documentation already in your database
• Table relationships - Identifying foreign keys and how tables connect

This metadata serves as the foundation for understanding your data structure. Unlike traditional BI tools where someone needs to manually define every field and relationship, Compass starts with what's already there and builds on it through use.

3. Contextual Learning

As your team asks questions and provides feedback, Compass builds a deeper understanding of your business. This happens organically through everyday use:

Business terminology and metrics:

• When someone asks about "MRR," Compass learns which tables and calculations define monthly recurring revenue for your organization
• Questions about "active users" teach Compass your specific criteria for what makes a user active
• Each metric definition is saved and reused for future questions

Table relationships in practice:

• Compass learns which joins matter for different types of questions
• It discovers which tables are commonly analyzed together
• It understands which fields are actually used vs. which exist but aren't relevant

Team-specific preferences:

• Different teams may care about different dimensions of the same data
• Sales might care about deal size while support cares about response time
• Compass adapts its answers based on who's asking and what context matters to them

4. Continuous Improvement

When Compass makes a mistake, you don't file a ticket or wait for someone to update a model—you correct it directly in the conversation.

How corrections work:

• If Compass includes the wrong data in an answer, tell it what's wrong
• If it misunderstands a business term, clarify what it actually means
• If it's missing important context, provide it in plain language

These corrections are automatically saved as context documents that Compass references for future questions. The next time someone asks a similar question, Compass will use your correction to provide a better answer.

This creates a virtuous cycle: the more your team uses Compass, the better it gets at understanding your specific business and data.

Learn more about Context Management.

Learning from Feedback

Compass improves its understanding based on how your team interacts with it:

Correcting Answers

When Compass provides an answer that's not quite right, you can correct it directly:

User: "Who are our top sales reps by win rate?"

Compass: Lists results including David Lee and Sarah Brown

User: "David Lee and Sarah Brown are not sales reps - please remove them."

Compass: Updates its understanding and remembers for future queries

Building Business Context

Over time, Compass learns:

• Business terminology - What "MRR", "churn", or "qualified lead" means in your organization
• Data relationships - How tables connect and which fields matter for different questions
• Team preferences - How different teams like to see their data presented

Context Documents

Your team's corrections and context are stored as context documents that Compass references when answering questions. These documents:

• Are version-controlled and auditable
• Can be reviewed and edited by administrators
• Build up over time as your team uses Compass
• Are organization-specific and secure

See Context Management for details on managing your organization's context.

Semantic Layer Integration

Coming Soon: For teams that already have a semantic layer (dbt semantic layer, Looker, Cube, etc.), we're building integrations to leverage your existing definitions.

If you've already invested in building out business logic in a semantic layer, Compass will be able to use those definitions alongside its learning capabilities.

Why This Approach Works

Traditional analytics tools create a bottleneck: someone needs to build and maintain a semantic layer before anyone can ask questions. This means weeks of upfront work, and every new metric or definition requires engineering effort.

Compass flips this model. Your team gets immediate access to data through natural language, and the system gets smarter as you use it. Every question asked, every correction made, and every bit of feedback provided strengthens Compass's understanding of your business.

The result? You get value from day one, and that value compounds over time as institutional knowledge builds up organically through everyday use.

See Getting Started to begin using Compass.