marivo 0.2.5

Python-native semantic modeling and analysis library for agentic analytics.

Marivo

Metric-centered analysis runtime for AI agents.

Marivo is a Python library — not a hosted service or a chat UI — that turns a data warehouse into something an AI agent can analyze reliably. It is not a text-to-SQL wrapper: Python declarations are the contract, ibis expressions are the execution language, and typed frames are the boundary between analysis steps.

• Unified semantics — Datasources, entities, metrics, dimensions, and relationships are declared in Python and addressed by semantic ref (sales.revenue), not raw table or column names.
• Metric-centered analysis — Sessions start from a trusted metric and chain typed intents (observe, compare, decompose, correlate, forecast), each returning a typed frame.
• Trustworthy evidence — Every operator records findings, propositions, and assessments, so any conclusion can be traced back to its inputs.
• Readiness gates — catalog.readiness() blocks incomplete semantic objects from analysis until authoring and access issues are resolved.

Requirements

Python 3.12 or newer.

Installation

pip install marivo

Install the backend extra that matches your datasource:

Backend	Install command
DuckDB	pip install "marivo[duckdb]"
MySQL	pip install "marivo[mysql]"
Postgres	pip install "marivo[postgres]"
ClickHouse	pip install "marivo[clickhouse]"
Trino	pip install "marivo[trino]"
All packaged backends	pip install "marivo[all]"

Marivo is deployed by installing the library where the agent runs, checking in the models/ declarations, and providing datasource secrets through environment variables referenced by *_env fields. There is no separate server.

Quick Start

Scaffold a project with the CLI:

marivo --version
marivo init

marivo init creates the project skeleton (marivo.toml, models/, .marivo/) and installs the marivo-semantic and marivo-analysis skills for Claude Code and Codex, so a coding agent can author the semantic layer with you.

Declare semantics under models/semantic/, then load and analyze:

import marivo.semantic as ms
import marivo.analysis as mv

catalog = ms.load()
report = catalog.readiness()
if report.status == "blocked":
    report.show()                       # resolve blockers before analysis

session = mv.session.get_or_create(name="revenue-check", question="Why did Q4 drop?")
revenue = session.catalog.get("metric.sales.revenue")
region = session.catalog.get("dimension.sales.orders.region")

current = session.observe(revenue, time_scope={"start": "2026-10-01", "end": "2027-01-01"}, grain="month", dimensions=[region])
baseline = session.observe(revenue, time_scope={"start": "2025-10-01", "end": "2026-01-01"}, grain="month", dimensions=[region])

delta = session.compare(current, baseline)
session.decompose(delta, axis=region).show()

In-project help is available at every surface: ms.help() and mv.help() list the surface; ms.help("metric") and mv.help("observe") drill into a symbol.

Documentation

Full guides live in the documentation site (built from site/):

• Installation — install, marivo init, deploy.
• Quick Start — author declarations, load a catalog, run a first analysis.
• Concepts — semantic layer, analysis workflow, readiness, and evidence.

Maintained agent guidance lives in marivo/skills/marivo-semantic and marivo/skills/marivo-analysis.

Development

python3 -m venv .venv
.venv/bin/pip install -e ".[dev,duckdb,trino]"

Repository commands run through the local virtualenv via make:

make lint
make typecheck
make examples-check
make test
make check

Read agent-guide.md before contributing. See CONTRIBUTING.md for the full workflow.