Skip to content

Core Concepts

The key ideas behind OnlyMetrix, explained without code.

Metrics

A metric is a named question about your data. Behind the scenes it's a SQL query, but you don't need to know SQL to use it. Your data team writes and maintains the queries — you just ask questions in plain English.

Example metrics:

  • total_revenue — "How much money have we made?"
  • customer_count — "How many customers do we have?"
  • churn_by_country — "Where are we losing customers?"

When you (or the AI agent) ask a question, OnlyMetrix finds the right metric and runs its pre-defined query. Same question, same answer, every time.

The metric catalog

All your metrics live in the Metric Store (Metrics page). Each metric has:

  • Name — unique identifier (e.g., total_revenue)
  • Description — what it measures in plain English
  • Tags — categories for organization (e.g., revenue, customers)
  • Tier — how important it is:
TierWhat it means
CoreBusiness-critical KPIs everyone sees
StandardRegular operational metrics
FoundationSupporting metrics, hidden by default

Datasources

OnlyMetrix connects to your existing data warehouse — PostgreSQL, ClickHouse, or Snowflake. No data is copied. Every query runs directly against your live tables.

The AI agent

The Dashboard chat is powered by an AI agent that:

  1. Understands your question — maps natural language to metrics
  2. Checks data reliability — warns you if data is stale or broken
  3. Runs queries — executes the right metrics with the right parameters
  4. Picks visualizations — area charts for trends, bar charts for breakdowns, tables for details
  5. Explains the results — narrative insights, not just raw numbers

The agent never writes SQL. It selects from your pre-defined metrics and runs them through OnlyMetrix's deterministic engine.

Reliability

OnlyMetrix continuously monitors your data for problems:

  • Stale data — a table hasn't been updated within its SLA
  • Schema drift — columns or tables have been renamed or deleted
  • Volume anomalies — row counts changed dramatically
  • Null spikes — a column suddenly has far more missing values

When the agent detects a reliability issue, it tells you before showing results — so you never make decisions on bad data without knowing it.

Analysis primitives

Beyond simple queries, OnlyMetrix has 22 analysis primitives — pre-built analytical methods like:

  • Root cause — "Why did revenue drop?"
  • Forecast — "Where is this metric heading?"
  • Scenario — "What if UK revenue grows 10%?"
  • Cohort — "How do user groups retain over time?"

The AI agent picks the right primitive based on your question. You don't need to know they exist — just ask naturally.

Structured vs. opaque metrics

When you create a metric, OnlyMetrix tries to compile it into an internal representation (IR). Compiled metrics are called structured — they support advanced features like dimension breakdowns, time granularity, and all 22 analysis primitives.

Some complex SQL can't be fully compiled. These are called opaque — they still work for basic queries but don't support advanced analysis. The Compiler tab on the Metrics page shows which metrics are structured vs. opaque.

For developers

If you prefer code over UI:

  • Python SDKpip install onlymetrixdocs
  • REST API — full CRUD for metrics, queries, analysis → docs
  • CLIomx terminal tool → docs
  • dbt/LookML import — bring existing metric definitions → docs

MIT Licensed (SDK) | Proprietary (Server)