Data platforms

Databricks

Unified analytics and lakehouse platform

Overview Context types Tools Use cases Connection Security

Overview

Modus walks Unity Catalog from catalog to column, mining the schemas you select so agents understand your lakehouse the way your platform team does.

What Modus reads

1Catalog
2Schema
3Table / View
4Column

You select the level Modus is allowed to mine. Everything below it is discovered automatically; everything outside it is never touched.

5context types mined

1agent tool groups

2connection methods

7security controls

Context types

Each object Modus mines produces typed context items, addressed by their position in the hierarchy. This is what an agent actually receives.

Schema metadata
The container Modus is scoped to: what lives inside it, how it is organised, and how much of it is actually used.
Table / View metadata
Names, descriptions, column lists, types, keys, and the join paths Modus observes between objects.
Table / View samples
A small, classified sample of representative rows so agents can tell an ISO date from an epoch, and a status enum from free text.
Column statistics
Distinct counts, null ratios, ranges, and top values per column, so an agent knows which column is the real grain.
Column samples
Representative values per column, redacted whenever classification flags them as sensitive.

Example context item

databricks_table_metadata

maingolddim_customer

Object: Delta table · 34 columns
Grain: One row per customer (SCD type 2)
Joins with: fct_orders.customer_id
Notes: Supersedes silver.customers_raw

Tools

Tools are granted per skill, never globally. Anything that changes state in Databricks is marked as a write and stops for human approval.

SQL Runner
Run governed read queries against the schemas a skill is scoped to. Statement categories (SELECT, INSERT, UPDATE, DELETE, CREATE) are granted per skill, so most agents get SELECT and nothing else.

Common use cases

Give agents the Unity Catalog view of what exists, without granting broad catalog access.
Explain which gold table supersedes which silver table.
Profile a table before an agent writes a query against 400 columns.

Connection method

Connect once. Modus stores credentials in a managed secret store, rotates what it can, and never asks an agent to handle them.

OAuth M2M service principal

A Databricks service principal with machine-to-machine OAuth, scoped by Unity Catalog grants.

Personal access token

A PAT bound to a read-only principal and a SQL warehouse you nominate.

Security & governance

Governance is applied once, at the layer every agent runs through, so it holds no matter which agent asks.

Metadata-first mining

Modus learns from structure and usage, not bulk data. Raw records stay in your environment; only the metadata needed to understand the system is stored.

Scoped selection

You choose exactly which schemas, folders, spaces, or boards Modus is allowed to see. Nothing outside the selection is ever mined.

Data classification

Sampled values are scanned for PII, financial, health, credential, and government-ID patterns, then classified as Sensitive, Flagged, or Non-sensitive. Your own overrides win over the defaults.

Group access by sensitivity

Map your identity groups to sensitivity tiers so each team only ever composes context at or below its clearance.

Read-only by default

Mining runs on a read-only credential. Any tool that writes back is a separate, explicitly granted action.

Query cache

Repeat read queries can be served from a governed cache with a configurable TTL, cutting warehouse cost and latency without bypassing access controls.

IP allow list

Modus connects from fixed egress IPs you allowlist in your network or cloud firewall, so the warehouse never has to accept traffic from the open internet.

Put Databricks to work.

Connect it in minutes and give every agent a governed understanding of what is inside, without moving a single row of data.

Book a demo