eatmydata
Local-First Question-to-SQL-to-Dashboard AI
Last verified:
What is eatmydata?
Eatmydata.ai is a local-first, privacy-focused AI analyst that lets you chat with your data using natural language questions. When you ask a question, the tool's agentic workflow produces multiple SQL queries that run directly in your browser against an in-browser SQLite database, then generates dashboard configuration code to visualize the results. The remote LLM never sees your actual data contents—all data is heavily sanitized, obfuscated, and auto-redacted for PII before any information is sent to the AI.
Key features include SQLite OPFS (Origin Private File System) adapted from wa-sqlite for local data querying, TurboQuant semantic indexing extension for SQLite with embeddings generation and vector search fully local, quantized PII detection models running in-browser, a QuickJS sandbox for executing AI-generated code, and an Orchestrator ↔ SQL Planner ↔ Coder agent loop that builds SQL queries and dashboards from user queries. Dashboards are rendered using Apache ECharts directly in your browser with no backend attached. The tool also includes a fork of xslx Community edition supporting styles for CSV/Excel file imports.
Eatmydata.ai is designed for data analysts, business users, developers, and anyone who needs quick data analytics answers under 10 seconds without wasting time debugging SQL or building charts manually. It's ideal for users who don't want to share personal customer Excel/CSV documents with external AI services like Claude or ChatGPT, or upload work-related datasets to cloud AI platforms. The tool handles datasets that might choke typical AI chatbots, including tiny 100k row files. It's fully open-sourced under MIT license as a static web app that you can run yourself.
The tool serves as a testbed for local-first AI projects, agentic workflows, and contextual data analysis experiments, but has grown into a daily-use tool for quick and dirty data analytics. It supports vector search for semantic queries (like finding addresses in 'Dallas' using similarity search) and formats output in Markdown for easy reading.
eatmydata pricing
Pricing model: Freemium
Free tier is all you need - the tool is free to use. It's fully open-sourced under MIT license as a static web app you can run yourself. You only pay for the remote LLM costs (like OpenRouter) when using cloud models - for example, gemini-3.1-flash-lite produced a query for $0.0033. There are no paid plans or subscription tiers from eatmydata.ai itself since it's MIT-licensed open source.
eatmydata pros
- Data never leaves your computer - fully browser-local execution
- Remote LLM never sees actual data contents
- Auto-redacts PII (personally identifiable information) on the fly
- Heavily sanitizes and obfuscates data at multiple points before LLM sends
- Gets answers under 10 seconds for simple data questions
- No time wasted debugging SQL manually
- No time wasted building charts for simple questions
- Works with datasets that choke Claude/ChatGPT (including 100k rows)
- Fully frontend app with no backend attached
- Open-source under MIT license - can run yourself
- SQLite OPFS for local data querying only
- TurboQuant semantic indexing with local embeddings generation
- Quantized PII detection models run entirely in browser
- QuickJS sandbox safely executes AI-generated code
- Orchestrator-SQL Planner-Coder agent loop automates query building
- Apache ECharts for rich dashboard visualization
- Supports CSV and Excel file imports with styles
- Semantic vector search for fuzzy matching (like finding 'Dallas')
- Markdown output formatting for easy reading
- Local semantic index for data indexing fully offline
eatmydata cons
- Requires remote LLM access (OpenRouter or other) - not 100% offline
- Only read-only access - can't modify original data files
- No backend means limited persistence beyond browser session
- Browser-based SQLite may have performance limits on very large datasets
- Some queries may fail depending on model choice (Dallas example error)
- Demo file selection window not scrollable on small screens
- No default model baked in yet - user must configure
- No default token quota implemented yet - user manages costs
- Primarily designed for quick analytics not complex enterprise workflows
- Limited to SQL-capable data structures
- No collaboration features for team data analysis
- Browser cleanup loses all data - not production-ready persistence
- May need technical knowledge to set up self-hosted version
- Vector search accuracy depends on model selection
- No mobile app - desktop browser only
- Currently a testbed project not enterprise-grade product
- Limited documentation beyond GitHub README
- No customer support channel mentioned
- Subscription costs depend on chosen LLM provider pricing
- Sembolic indexing adds computational overhead in browser
Frequently asked questions about eatmydata
What is Eatmydata.ai?
Eatmydata.ai is a local-first, privacy-focused AI analyst that lets you chat with your data using natural language. You ask a question, agents produce multiple SQL queries to in-browser SQLite (never seeing results), and write dashboard configuration code. The remote LLM never sees the contents of data it analyzes because all data is heavily sanitized and obfuscated at several points.
How does Eatmydata.ai keep my data private?
All data sent to LLMs is heavily sanitized and obfuscated at several points. The tool includes quantized PII detection models running straight in the browser that auto-redact personally identifiable information on the fly. The remote LLM never sees the actual contents of your data - it only sees sanitized, obfuscated queries and schema information.
What data formats does Eatmydata.ai support?
The tool supports CSV and Excel file imports. It includes a fork of xslx Community edition that supports styles (which is missing in the OSS version upstream). You can import your data files directly into the browser-based SQLite database.
How fast does Eatmydata.ai provide answers?
The tool is designed for quick data analytics when you literally need an answer under 10 seconds. It automates SQL query generation and dashboard creation so you don't waste time debugging SQL or building charts for simple data questions.
Can I run Eatmydata.ai myself without using the website?
Yes, Eatmydata.ai is fully open-sourced under MIT license at https://github.com/eatmydata-org/eatmydata. It's a static web app that you can run yourself locally without using the hosted website.
What happens if my query fails or produces an error?
Some queries may fail depending on the model choice. For example, the 'show me how many people are from Dallas' query in AdventureWorks LT threw an error with some models. However, with gemini-3.1-flash-lite for all agents, it produces the correct SQL from the first try. The author plans to bake in default models and more examples later.
Does Eatmydata.ai upload my data to any cloud service?
No, Eatmydata.ai works fully in your browser without uploading data to the cloud. It uses SQLite OPFS (Origin Private File System) adapted from wa-sqlite, so data is queried only locally. This is why it's called 'local-first' - the data never leaves your computer.
What LLM providers does Eatmydata.ai support?
The tool uses OpenRouter or other remote LLM providers. The author mentions using gemini-3.1-flash-lite for all agents, which produced SQL for $0.0033. You can choose different models depending on your needs and cost preferences.
Can Eatmydata.ai handle large datasets?
Yes, the tool handles datasets that often choke Claude or ChatGPT, including tiny 100k row files. The author specifically mentioned not wanting to upload work-related datasets to external AI services because they often choke on this size of data.
What visualization library does Eatmydata.ai use for dashboards?
Eatmydata.ai uses Apache ECharts for dashboards. The sandboxed QuickJS runs the AI-generated dashboard configuration code to produce rich dashboards directly in your browser with no backend attached.