Evidence Core Review: BI as Code

What Is Evidence Core?

Evidence Core is an open-source business intelligence framework that treats analytics as code. It is the same engine that powers Evidence Studio, open sourced on August 26, 2026, so any team can run a commercial-grade BI stack on its own infrastructure. Instead of drag-and-drop interfaces, you write SQL queries and Markdown page layouts in a Git repository, and the Evidence CLI compiles them into fast, interactive data apps.

Live queries are central to the experience: when someone opens a deployed project, its components query the underlying database in real time, so dashboards never go stale between builds. Research teams, data analysts, and product engineers use the framework because it brings version control, code review, and CI/CD to reporting — the same discipline developers already apply to application code. Projects build into Docker containers with basic authentication baked in, which means hosting is just another container deployment. The whole stack stays portable: a team can start in the open-source project, prove out a reporting layer, and push the same artifacts to a hosted environment when it needs a control plane with access control and scheduled email delivery.

The Evidence Analytics Agent

The open-source project is the framework; the Evidence Analytics Agent is the AI assistant that ships with Evidence Studio and works against the projects you build with it. The agent reads every dashboard, report, and data source in a project and answers natural-language data questions in plain language. It can also build new insights and save them back to the project, where teammates review, branch, and iterate on the results.

The agent is available in Evidence Studio, in Slack, and through any MCP client such as Claude Desktop or ChatGPT, and it respects the same row-level security and page-level access controls already configured for a project. Teams add custom context and skills to tailor the agent to their business, while built-in evals and observability track how well the assistant performs over time. Because the assistant operates on the same SQL and Markdown artifacts that the team already reviews, every insight it saves lands in version control like any other change — so nothing it produces is a black box.

• Answers data questions using every report in the project as context
• Builds and saves new insights that the team can review and branch
• Works in Evidence Studio, Slack, and MCP clients like Claude Desktop and ChatGPT
• Honors row-level security and page-level access controls

Live Queries and One-Time Metrics

A headline change in the new framework is live queries. Formerly, refreshing a deployed report meant running a build; now the components of a deployed project query the underlying database directly, so the data a viewer sees is always current. This removes an entire class of stale-data problems and makes self-hosted deployments feel like a live product rather than a static export. A single deployment can serve an entire company with the freshest data possible, using whatever warehouse already powers the business.

The release also introduces standalone, one-time metrics. A calculation such as revenue can be defined once in a YAML file and referenced from every chart, table, and summary on the site. If the definition changes, you update it in a single place instead of hunting through every component that uses it, which keeps definitions consistent and auditable across an organization.

Finally, the new syntax is terser and more declarative. Components manage their own queries and aggregations, projects contain far less code for the same output, and the markup can be validated statically. Coding agents get a specific error when they write invalid markup instead of starting the project and trying to infer what broke.

• Deployed dashboards hit your warehouse on every load
• Metrics defined once in YAML and reused across all components
• Terser, declarative syntax that validates statically
• Standalone CLI bundles dev server, validator, and production server

Self-Hosting and Deployment

Getting started is a one-liner on macOS, Linux, and Windows, after which the CLI scaffolds a project with evidence init and starts a local preview with evidence dev. Everything a developer needs — the development server, validator, production server, and documentation — ships in a single binary, so there is no dependency chain to maintain. The binary approach also makes the tool friendly to ephemeral CI environments, where a pipeline can validate an entire project without installing a toolchain.

Publishing is just as simple. Projects build into Docker containers that include basic authentication, so they run on any container host, including a company's own data center or a private VPC. A single CLI command publishes the same project to Evidence Studio for teams that want a hosted environment with a browser IDE, email distribution, and the agent. Existing users of the earlier open-source project can run a migrate command that handles most of the move to Evidence Core, which keeps upgrade paths short.

• Docker containers with built-in basic authentication
• Deploy anywhere you host containers, including private VPCs
• One CLI command to publish to Evidence Studio
• migrate command smooths the path from the older Evidence project

Evidence Core vs Tableau vs Looker

Tableau and Looker are enterprise BI platforms with drag-and-drop interfaces, visual query builders, and large support ecosystems. They fit analysts who prefer GUI tools and organizations that want an established vendor with enterprise contracts and a deep bench of certified professionals.

The framework takes the opposite route: code-first, Git-native, and purpose-built for coding agents. It is free to self-host, fast to iterate on, and produces dashboards that agents can read, edit, and validate automatically. The trade-off is SQL comfort — there is no visual query builder for non-technical authors, and the community around the new framework is younger than those of the incumbents.

• Tableau — visual drag-and-drop BI versus code-first Evidence Core
• Looker — LookML semantics versus plain SQL and Markdown
• Free to self-host versus Tableau and Looker paid licenses
• Built for coding-agent workflows none of the incumbents offer

Evidence Core Pricing

Open source and self-hosted free forever; Evidence Studio from $2,500/mo (Team, unlimited users, 30-day trial) and Enterprise at custom pricing

Best For

Recommended use cases and scenarios where Evidence Core shines.

Frequently Asked Questions

Common questions about Evidence Core, answered.