Metric Reliability Infrastructure

Know which decisions you can't trust.

OnlyMetrix traces data quality from source table to business decision. Not just which table is broken — which board decision depends on it.

Works on Snowflake, Postgres, and ClickHouse. Imports from dbt. No dbt required.

Get Started GitHub

omx reliability affected-by invoices

Affected metrics — 5

total_revenue 95% → 70%

avg_order_value: 97% → 72%

Critical "Q1 Budget Reallocation" — board decision, 3d ago

5 metrics 2 dashboards 1 decision

The Core Difference

Two approaches to data reliability

Observability tools

Table-level monitoring

Table breaks

Alert generated

row count changed
column has nulls
freshness violated

Data team investigates manually

Impact unknown

You know the table is broken

You don't know what decisions depend on it.

OnlyMetrix

Decision-level tracing

Table breaks

Compiler traces the IR

table → column → metric
metric → dashboard
metric → decision

Severity ranked

CRITICAL: decision can't be trusted

Auto-inferred tests from the IR

Drift detection on source tables

You know exactly which decisions to distrust.

Monte Carlo tells you the table is broken. OnlyMetrix tells you which decisions you can't trust.

The Problem

Data breaks. You find out from the CFO.

Every data team knows the feeling:

A pipeline fails silently. A table goes stale.
A metric returns wrong numbers for days.
Someone makes a board decision based on broken data.

Existing observability tools tell you:

which table is broken
which column has nulls
which pipeline failed

They don't know which business decisions depend on it.

invoices table goes stale

total_revenue reliability 95% → 70%

Revenue Overview dashboard stale data

Q1 Budget Realloc. decision can't trust

One table. Three layers of downstream impact.

How It Works

Three layers. One platform.

Most tools operate at one layer. OnlyMetrix spans all three — and the layers compound.

Compiler + IR

Compiles metrics into an intermediate representation: source tables, aggregation logic, join paths, time columns, downstream dependencies.

Validation + Drift

Auto-inferred tests from the IR. No YAML. SUM implies non-negative. time_column implies freshness. Plus volume, null rate, and schema drift detection.

Reasoning + Agents

Deterministic investigation agents. The LLM picks which commands to run. The Rust engine executes. Every action is signed.

Auto-inferred tests

6 test rules derived from the IR. No config files. No YAML. The compiler figures it out.

Drift detection

Volume, null rates, mean shifts, cardinality, schema changes. Baselines established automatically.

Decision tracing

Every business decision linked to the metrics it depends on. When data breaks, you know which decisions to distrust.

Built for production

Rust core. 15-minute reliability cycle. Compound reliability scoring. Sub-millisecond overhead.

Developer-first

Infrastructure you can pipe, script, and automate

40+ CLI commands. Full Python SDK. dbt integration via manifest.json. MCP server for Claude and Cursor. Every command returns structured JSON.

omx reliability check

Overall metric health, right now. See which metrics are healthy, degraded, or unreliable.

omx reliability affected-by

What breaks when a table breaks. Traces the full chain: table → metric → dashboard → decision.

omx validation run

Run compiler-inferred tests. 6 rules, zero config. See what tests were inferred and why.

omx dbt sync

Read manifest.json, compile your dbt models into the IR. One command.

omx agent "..."

Deterministic investigation. The LLM picks commands. The Rust engine executes. Every action audited.

Integrations

Works With Your Stack

Connect your warehouse, install the SDK, or deploy as an MCP server. OnlyMetrix fits where you already are.

Snowflake

live

Native Snowflake connector with session management and token refresh

PostgreSQL

live

Connection pooling, read-only enforcement, statement timeouts

ClickHouse

live

HTTP API integration with secure and standard modes

Python SDK

live

pip install onlymetrix - query metrics from any Python app

MCP Protocol

live

Connect any MCP-compatible agent - Claude, GPT, custom agents

dbt

live

Sync metric definitions and column descriptions from dbt manifests

Where OnlyMetrix fits in your stack.

OnlyMetrix does

Metric compiler + IR
Auto-inferred data tests
Drift detection
Table → decision tracing

Other tools stop at

Table-level monitoring
Manual test configuration
Row count alerts
No downstream impact

OnlyMetrix complements dbt — it doesn't replace it. Your transformation layer stays in dbt. Your reliability layer lives here.

Under the Hood

24 tests inferred. Zero written by you.

The compiler reads your metric IR and infers validation tests automatically. No YAML. No config.

$ omx validation run

# 24 tests inferred. 0 written by you.

✓ total_revenue  non-negative · HIGH
✓ total_revenue  freshness · HIGH
✓ churn_rate     bounded 0–1 · HIGH
✓ revenue_by_country sum consistency

✗ total_revenue  non-negative · FAILING

FAILURE DETAIL
  severity      HIGH
  rows affected  1,847 (2.3% of 80,300)
  first seen    today (new failure)
  inspect
    SELECT * FROM invoices
    WHERE total_amount < 0 LIMIT 10

The compiler infers the tests

SUM aggregation → non-negative check. time_column → freshness check. COUNT/COUNT → bounded 0-1. Join paths → referential integrity. No configuration required.

Agent Response

Tests inferred 24

Passing 23

Failing 1

Reliability impact total_revenue → 75%

Decisions at risk 1 (Q1 Budget)

Proof Points

Real numbers, not promises

632 tests, zero failures

6 auto-inferred validation rules

5 drift detection types

15min reliability scan cycle

40+ CLI commands, structured JSON

Dashboard Canvas

Dashboards built from governed metrics

Every chart, KPI, and table pulls from the compiled IR. Reliability badges are live. Click any card to trace to source.

Quarterly Revenue

$686,654

Total Customers

5.9K

Invoice Count

44.9K

Average Amount

370.98

Quarterly Revenue Trend

Oct 2009Jan 2010Apr 2010Jul 2010Oct 2010Jan 2011

Comments Activity Decisions

Data Science · 2h ago

Revenue spike in Q3 — is this real or a data issue?

Reply Decision

OnlyMetrix · 1h ago

Verified. Reliability 95%. No anomalies detected in invoices table.

Investigation Pipeline

Visual investigation from source to insight

Drag metrics, analysis primitives, and filters onto a canvas. Connect them. Run the pipeline. Get structured findings.

IR Explorer

See the dependency graph. Know what breaks.

Every metric traces back to source tables. Every decision traces forward to the data it depends on. Reliability scores propagate through the graph.

Dependency graph Semantic tests Impact analysis

total_revenue

Metric · compiled IR

PROPERTIES

Reliability95%

Source tableinvoices

AggregationSUM(amount)

Last refreshed2h ago

AUTO-GENERATED TESTS

✓Non-negative values

✓Breakdown consistency

!Freshness check

Built by someone who spent years as a data analyst firefighting broken dashboards with no way to trace which upstream table caused the issue.

Know which decisions you can trust.

Connect your warehouse. The compiler infers tests. You see which decisions to distrust — automatically.

Get Started GitHub