Semantics
What does the metric mean?
A Data Analysis Harness for AI Agents
Marivo is the runtime layer between an AI agent and business data. It keeps business meaning, analytical steps, session state, and evidence connected, so findings used in operations, governance, and decision-making remain inspectable, resumable, and traceable.
Explore Marivo's design and practice01 / Analysis path
The agent chooses how to investigate. Marivo keeps each step within explicit, verifiable boundaries.
Why did revenue decline last quarter, and which region should we investigate first?
What does the metric mean?
Which step is valid next?
What has already been done?
What supports the finding?
Findings remain attached to their evidence, scope, and limitations.
02 / Four foundations
Define metrics, dimensions, relationships, and guardrails once as reusable contracts rather than reconstructing them in every analysis.
Semantic LayerThe agent chooses what to do next, while typed inputs, outputs, and failure modes keep the work visible and reviewable.
Typed Analysis DSLQuestions, intermediate results, and analytical artifacts remain together in a persistent session that can be resumed and extended.
Analysis SessionMaterial findings remain linked to source results, analytical scope, and limitations, making trust and traceability part of the handoff.
EvidenceMarivo's design philosophy
An error in personal exploration may cost an hour. In an enterprise, analysis drives operations, governance, and decisions. Once findings carry business consequences, trust and traceability are not optional—they are prerequisites for action.
Business definitions are explicit
The investigation is resumable
Every finding is traceable
03 / Start here
Install Marivo, prepare a semantic layer, and give an agent a business question to investigate within explicit boundaries.
Read the docs →BlogExplore product design, analytical practice, release notes, and community updates.
Read the blog →GitHubExplore the implementation, open an issue, or help improve the infrastructure for agent-driven data analysis.
Open GitHub ↗