Graphene: Enable AI agents to write analytics code 10x faster with everything-as-code

Graphene, an open-source analytics framework, enables AI coding agents to generate accurate data reports and dashboards 10x faster. By combining SQL-based governance with full query capability, the project lets agents write analytics code directly into repositories, avoiding the YAML verbosity that slows agent iteration. Now on GitHub with 207 stars, Graphene targets a critical gap: agents need a token-efficient way to work with data.

What Graphene does

Graphene is an everything-as-code analytics platform that treats data models, queries, dashboards, and visualizations as code—versioned in Git and deployable in pull requests. Rather than forcing agents through GUI workflows, Graphene gives them:

• Graphene SQL: A semantic layer combining SQL governance with full ANSI SQL capability. Macros define metrics deterministically; queries reference tables and measures via dot notation.
• Markdown dashboards: .md files define pages with inlined SQL queries and visualization components (HTML, CSS, JavaScript, ECharts).
• Database support: Snowflake, BigQuery, ClickHouse, Postgres, MotherDuck, and DuckDB.
• Chunked semantic models: .gsql files define tables, dimensions, and measures; modeled relationships enable agent traversal.

Why agents need this

Building analytics in traditional BI tools is expensive for agents: click workflows don't scale, YAML config bloats prompts, and schema mismatches halt iteration. Graphene solves this by:

• Reducing token overhead vs. YAML-based configurations
• Letting agents read and write analytics code end-to-end
• Enabling dashboard creation in the same PR that changes application code
• Validating queries against live schema without query execution

A typical agent workflow: read the semantic model (Graphene SQL), generate a dashboard query, write the .md page with inline SQL, and commit—no backend round-trips, no serialization overhead.

Availability and status

Graphene is open source under the Elastic License 2.0 (free for internal use; commercial applications require permission). The project shipped recently and is actively maintained. Installation via npm: available on PyPI and as an npm package, compatible with pnpm and yarn.

Who this serves

• Autonomous coding agents performing data analytics
• Teams embedding agents in monorepos with analytics needs
• Organizations wanting agent-driven business intelligence without BI tool sprawl
• Developers integrating analytics into application code

Graphene doesn't replace traditional BI platforms; it targets the agent-shaped gap where traditional tooling blocks iteration.

Why it matters

As AI agents become responsible for data-driven decisions, the friction of traditional analytics—even automated—becomes a bottleneck. Graphene removes that friction by treating analytics like code: write it, test it, version it, review it. For platform teams deploying agents, Graphene means agents can own analytics end-to-end, reducing the need for humans to manually define dashboards or sanitize agent-generated SQL. The 10x speedup claim reflects token efficiency and fewer round-trips—both measure an agent's ability to complete analytics tasks in a single coherent effort.

Expect more "everything-as-code" tools for agents; Graphene is the first to make that work for analytics at scale.