Rath: Automated Exploratory Data Analysis with Causal Discovery

What Rath Does and Who It Is For

Rath is a web application that applies an augmented analytic engine to a loaded dataset and returns insight recommendations, pattern detection results, and automatically generated visualizations. The README describes it as more than an open-source alternative to data analysis tools like Tableau: it automates the exploratory phase rather than requiring a user to manually specify chart types, dimensions, and measures.

The intended users are data scientists, business analysts, and engineers who work with tabular data and want to accelerate the exploration phase. Rath covers the pipeline from data import through visual analysis. It accepts data from CSV and JSON files as well as online databases. It includes a causal analysis module for identifying relationships between variables, which separates it from pure visualization tools.

A sandbox demo is available at https://rath.kanaries.net for testing parts of the feature set without installing anything. The README describes Rath as being in open alpha, which means the code and documentation are actively evolving and stable production deployments are not guaranteed.

Running Rath Locally with Yarn or Docker

Rath is a monorepo managed with Yarn workspaces. Node.js 24.x is required, as specified in package.json. After cloning the repository, the frontend client build runs with:

yarn workspace rath-client build

To start the frontend development server:

yarn workspace rath-client start

For the backend service in development mode:

yarn workspace backend dev

The Docker Compose configuration provides the simplest path to a running instance. The docker-compose.yml defines two services: base, which runs the compiled client on port 9083, and connector-api, which runs the database connector backend on the same network as the base service. Running both with Docker Compose exposes the full application at port 9083.

The package.json shows that the package manager is Yarn 1.22.22 and Node.js 24.x is required. Several packages are hoisted with the nohoist configuration, including vega, vega-lite, vega-embed, and visual-insights, which means they are installed locally in their workspace rather than at the repository root.

For teams who want only the manual drag-and-drop chart exploration module without the full Rath application, the README describes Graphic Walker as an independent embeddable module. Install it in a JavaScript project with:

npm i --save @kanaries/graphic-walker

Or with Yarn:

yarn add @kanaries/graphic-walker

AutoPilot, AutoVis, and the Recommendation Engine

The AutoPilot feature is Rath's primary differentiator. It runs the augmented analytic engine against the loaded dataset and returns insight recommendations with supporting visualizations after a single click. The README states that it discovers patterns, insights, and causals automatically. The documentation describes the engine as generating visualizations that minimize visual perception error, though the README does not elaborate on the specific algorithm used.

AutoVis is a lower-level component of the same pipeline. It generates what the README describes as the best visualization for a selected set of data columns, focusing the user's attention on choosing dimensions and variables rather than deciding on chart type or encoding.

The Copilot mode is a semi-automated alternative to AutoPilot. In Copilot mode, Rath learns from the user's interactions and generates relevant recommendations based on observed intent, rather than running a fully automated scan. The README contrasts this as a middle ground between full automation and fully manual exploration.

The natural language interface integrates with GPT to answer questions about the loaded dataset and return answers as visualizations. The README presents this as an additional input method for users who prefer to describe what they want to see in text rather than configuring chart parameters.

Causal Analysis, Data Painter, and the Less-Documented Modules

Rath's causal analysis module covers four areas: causal discovery from the dataset, editable graphical causal models, causal interpretability tools, and what-if analysis. The README states this can help create better prediction models and support business decisions. The causal module is described as in alpha stage, meaning it is present but may have incomplete functionality or limited documentation.

The Data Painter is an interactive mode where the user brushes or colors data points directly on a visualization. The README describes it as instinctive and powerful for identifying clusters and anomalies that automated detection might miss. A demonstration video is linked from the README. The tool is positioned for users who prefer direct manipulation over algorithmic output.

The Data Wrangler handles data preparation. The README describes it as automated, generating suggestions for transformations and cleaning. It lists capabilities including summary statistics, predictive transformation operations, and cleaning suggestions. The README does not document the specific transformation types available.

The Dashboard module allows building multi-chart views with an automated designer that suggests dashboard layouts. The README does not specify the output formats or sharing mechanisms available from the dashboard builder.

The AGPL-3.0 License and What It Means for Deployment

Rath is released under the GNU Affero General Public License version 3.0. AGPL is a strong copyleft license. The key difference from the standard GPL is the network service clause: if a modified version of Rath is run as a service accessible to users over a network, the operator must make the complete source code of the modified version available to those users under AGPL terms.

This has direct consequences for commercial deployment. A company that adds features to Rath and runs it as an internal analytics service accessible over the company network must publish those modifications under AGPL. A company that offers Rath-based analytics as a paid service to external customers must do the same.

For internal use where the software is not accessible over a network to third parties, AGPL behaves similarly to GPL. An engineer who forks Rath for personal use or for a closed network with no outside access is not required to publish modifications.

The README does not mention a commercial license option. Organizations whose open-source policies prohibit AGPL software in network services should treat Rath as incompatible with that policy without first evaluating whether a commercial agreement is available from Kanaries.

Comparing with Tableau and Other Business Intelligence Tools

Tableau is a commercial visual analytics platform that provides a drag-and-drop interface for building charts and dashboards from database connections and file sources. It is a closed-source SaaS and desktop product with a subscription or perpetual license. The fundamental difference in approach is automation: Tableau requires the analyst to specify what to visualize; Rath's AutoPilot runs the analysis and returns recommendations without that specification step.

For a data scientist already writing analysis code, Tableau adds a visual layer on top of existing data but does not automate the insight discovery phase. Rath inverts this: it starts from discovery and provides interactive controls for refinement.

The AGPL license and open alpha status position Rath differently from commercial BI tools. An engineering team that wants to embed data exploration in their own product will face the AGPL network clause when using Rath's server components. The Graphic Walker component, installable independently as @kanaries/graphic-walker, offers manual drag-and-drop chart exploration with a less restrictive embedding footprint, though it does not include AutoPilot or causal analysis.

Open Alpha Status, Recent Activity, and the runcell Companion

The README explicitly states that Rath is in open alpha stage, with the team working on improving code and documentation. The latest GitHub releases are from 2023: versions 2.1.0, 2.0.0, and the v2.0 beta. The package.json version field reads 1.1.0. The last push to the repository was on 2026-08-14.

The README also mentions runcell, a separate product from the same team. The README describes runcell as an AI Code Agent in Jupyter that understands code, data, and cells, and can generate code, execute cells, and take actions. It is installable with `pip install runcell` and is available at runcell.dev. This is a distinct product from Rath and is mentioned in the README as an example of the team's related work.

The repository includes smoke tests using Playwright and a frontend test suite that runs with `yarn workspace rath-client test`. The README invites users to report bugs and feature requests through GitHub issues, and to contact [email protected] for feedback about current use cases.

Editorial conclusion

Rath is worth evaluating for data scientists and analytics engineers who want automated insight discovery without writing chart code, and who can run a local or self-hosted instance. The AGPL-3.0 license is a genuine constraint: any organization that deploys Rath as a network service must release their modifications publicly. The open alpha status means the API and feature set are not stable. Verify that Rath's causal analysis and AutoPilot features cover the workflows you need before committing to self-hosting it, and check AGPL compatibility with your organization's open-source policy before using it in a production service.

Frequently asked questions

Official sources

1. Kanaries/Rath on GitHub
2. License: AGPL-3.0
3. Project website
4. README
5. Releases