Open-source AI analyst for your database — adapter-first, self-hostable, and built for teams.

Ask your database questions.
Keep control of the queries.

Helium lets you connect your database, ask questions in plain English, inspect generated queries, and get trusted answers from your own data. Start with MySQL today, extend to any database tomorrow.

Open SourceMulti-databaseSelf-hostableQuery PreviewRead-only ModeAdapter-first
helium — localhost:3000
Your question

Which customers generated the most revenue last month?

Generated Query
ValidatedRead-only
SELECT
  c.name,
  SUM(o.total_amount) AS revenue
FROM customers c
JOIN orders o ON o.customer_id = c.id
WHERE o.created_at >= DATE_SUB(CURDATE(), INTERVAL 1 MONTH)
GROUP BY c.name
ORDER BY revenue DESC
LIMIT 10;
MySQLQuery Preview
Results3 rows · 12ms
CustomerRevenueOrdersLast Order
Acme Retail$42,830128Apr 29
Northstar Labs$31,42096Apr 28
BluePeak Co.$18,90551Apr 26
Explanation

This query joins the customers and orders tables, filters for orders created in the last month, groups by customer name, sums the total amount as revenue, and returns the top 10 customers sorted by revenue descending.

MIT Licensed
Multi-database
Adapter-first
Built in Public
Self-hostable
Query Preview
The problem

Your database is powerful. Your team should not need to wait three days for every answer.

Dashboards answer known questions. Helium helps you explore the unknown ones.

Data-rich, analyst-poor

Teams have more data than ever but not enough analysts to answer every question.

Founders need fast answers

Operators cannot wait three days for a report. Decisions move faster than dashboards.

Analysts are the bottleneck

Every ad-hoc question becomes a ticket. Analysts become query concierges, not strategists.

Dashboards answer yesterday

Pre-built dashboards answer the questions you knew to ask last quarter.

AI tools lack context

Generic AI chatbots do not understand your schema, joins, metrics, or permissions.

Direct access is risky

Without query preview, validation, and read-only controls, database access is dangerous.

Features

Ask in English. Review the query. Trust the answer.

Generated queries should be visible, explainable, and reviewable. Helium is building the full analyst stack, one layer at a time.

Available

Natural-language questions

Ask questions in plain English and turn them into database queries.

Available

MySQL adapter

Start with a production-quality MySQL adapter.

Available

Query preview

See generated queries before anything runs against your database.

Available

Read-only execution

Encourage safe database credentials and read-only queries.

Available

Schema understanding

Read tables, collections, columns, relationships, and metadata automatically.

Available

Result tables

Display clean, structured tabular answers from your data.

Available

Query explanation

Explain what the generated query is doing in plain English.

Available

Query validation layer

Check queries for correctness and safety before execution.

Available

Adapter-first architecture

Every database integration follows a clean adapter contract.

Available

Self-hosted deployment

Keep data and credentials inside your own environment.

Coming next

PostgreSQL adapter

Add Postgres support through the same adapter interface.

Coming next

Saved questions

Save and reuse recurring business questions.

Coming next

Query history

Track past prompts, queries, and results for reference.

Coming next

Chart generation

Turn result sets into visual charts automatically.

Coming next

CSV export

Export result tables for downstream analysis.

Coming next

Semantic metrics

Define revenue, churn, active users, GMV, retention, and more.

Coming next

Business glossary

Map business terms to database fields for better accuracy.

Coming next

Prompt context memory

Remember useful schema and query patterns across sessions.

Coming next

Error-aware retries

If a query fails, explain the issue and suggest a corrected version.

Coming next

Relationship discovery

Help the model understand how your tables and collections connect.

Roadmap

Dashboard builder

Create lightweight dashboards from saved questions.

Roadmap

Role-based access

Control who can ask what, at the question and table level.

Roadmap

Audit logs

Track questions, queries, execution time, and user activity.

Roadmap

Approval workflows

Require review before running sensitive queries.

Roadmap

dbt integration

Use dbt models and metrics as trusted context.

Roadmap

Slack / Teams bot

Ask database questions from your team chat.

Roadmap

API access

Use Helium as an analyst API in internal tools.

Roadmap

PII detection

Warn before exposing sensitive columns in results.

Roadmap

Model provider flexibility

Support OpenAI, Anthropic, local models, and other LLM providers.

Roadmap

On-prem enterprise mode

Run Helium fully inside private infrastructure.

How it works

From question to answer, safely.

Every step is visible. Every query is reviewed before execution. No hidden operations.

01

Connect database

Point Helium at your database with read-only credentials.

02

Introspect schema

Helium reads your tables, collections, types, and relationships.

03

Ask a question

Type a question in plain English about your data.

04

Generate query

The analyst engine creates a native query from your question.

05

Preview & validate

Review the generated query before it touches your database.

06

Run read-only query

Execute the validated query safely against your database.

07

View results

See clean tabular results with an explanation of the answer.

08

Save or export

Save the question, export to CSV, or generate a chart.

Architecture

Built with safety before execution.

Every database is an adapter. New adapters implement a common interface. Safety and validation sit before execution. Semantic context improves accuracy over time.

User Question01
Analyst Engine02
Schema Context Store03
Prompt Builder04
Query Generator05
Query Validator06
Policy / Safety Layer07
Database Adapter08
Query Executor09
Result Formatter10
Explanation Layer11
Adapters

Every database is just another adapter.

Helium is designed so every database integration follows a clean contract. Connect to MySQL, PostgreSQL, MongoDB, and more through adapters.

DatabaseAdapter Interface
interface DatabaseAdapter {
  connect(config: AdapterConfig): Promise<Connection>
  introspectSchema(): Promise<Schema>
  validateQuery(query: string): ValidationResult
  executeReadOnly(query: string): Promise<ResultSet>
  explainQuery(query: string): Promise<Explanation>
}
MySQL
Available now
PostgreSQL
Next
SQLite
Planned
BigQuery
Planned
Snowflake
Planned
ClickHouse
Planned
DuckDB
Planned
MongoDB
Experimental
Security

Designed for safety. Not an afterthought.

Business users need answers, but engineering teams need safety. Helium is designed so both can coexist.

Active

Self-hostable by design

Run Helium entirely inside your own infrastructure.

Active

Credentials stay local

Database credentials never leave your environment.

Active

Read-only database users

Designed for read-only database credentials.

Active

Query preview before execution

Every generated query is shown before it runs.

Active

Query validation

Queries are validated for safety before execution.

Active

No hidden queries

No query runs without explicit user action.

Active

No training on your data

Your data is not used for model training by default.

Roadmap

Sensitive column warnings

Detect and warn before exposing sensitive columns.

Roadmap

Audit logs

Track every question, query, and execution.

Roadmap

Role-based access control

Control who can ask what questions.

Roadmap

Approval workflows

Require review before running sensitive queries.

Open Source

Open-source because the analyst layer should belong to the teams running the data.

MIT-licensed, contributor-friendly, and built in public. The best data infrastructure is shaped by the people who use it.

MIT License

Free to use, modify, and distribute.

Built in public

Development happens in the open on GitHub.

Contributor-friendly

Clean adapter interface makes contributing straightforward.

High code standards

TypeScript-first, tested, linted, formatted.

Clear docs

Architecture and adapter guides for new contributors.

Testable architecture

Modular design that is easy to test and extend.

Roadmap shaped by users

Features prioritized by community feedback.

Good first issues

Tagged issues to help new contributors get started.

Roadmap

Ambitious vision, honest timeline.

We are building Helium one layer at a time. Here is what is available, what is next, and what is on the horizon.

Now

Foundation

Landing page
MySQL adapter foundation
Schema introspection
Natural language to queries
Query preview
Read-only query execution
Result tables

Next

Expand & improve

PostgreSQL adapter
Saved questions
Query history
CSV export
Chart generation
Semantic metric definitions
Business glossary
Query correction loop

Later

Scale & enterprise

Dashboard builder
Slack / Teams bot
Audit logs
Role-based access
Approval workflows
dbt integration
Multi-tenant workspaces
Adapter marketplace
Evaluation suite
On-prem deployment mode
Use cases

Built for every team that needs answers.

Different roles, same need: fast, trusted answers from the database without waiting on analysts or query experts.

Founder

Which customers, products, or channels are driving revenue?

Operations

Where are orders, payments, refunds, or shipments getting stuck?

Product

What features are users adopting, ignoring, or dropping off from?

Support

What customer patterns explain repeated complaints?

Finance

What changed in revenue, refunds, margins, or unpaid invoices?

Data Analyst

Generate first-draft queries faster, then inspect and improve them.

Engineering

Give internal teams safer access to analytical answers.

Growth

Find cohorts, conversion drops, and campaign performance insights.

Developer experience

Engineers deserve clean infrastructure.

Helium is built with the same standards you expect from the tools you already trust.

Clean TypeScript codebase

Strict typing across the entire project.

Adapter contracts

Typed interfaces for every database adapter.

Modular packages

Organized into focused, composable modules.

Environment-based config

Configuration through environment variables.

Test-first adapters

Database adapters ship with test coverage.

Seeded demo database

Get started with a pre-populated demo dataset.

Local dev setup

Start developing with a single command.

Roadmap

Docker support

Containerized deployment support on the roadmap.

Roadmap

CI checks

Automated testing and linting pipeline planned.

Linting & formatting

ESLint and Prettier configured for consistency.

Error boundaries

Graceful error handling across the stack.

Roadmap

Observability hooks

Planned instrumentation for monitoring.

Comparison

How Helium fits in.

Generic AI chat is flexible but unsafe. Traditional BI is great for recurring metrics but slow for new questions. Helium is for safe, inspectable, open-source ad-hoc analysis.

CapabilityGeneric AI ChatTraditional BIHelium
Understands live schema
Shows query before running
Self-hostable
Adapter-based
Open-source
Good for ad-hoc questions
Good for recurring dashboards
Developer-extensible
Safety controls

Start asking your database questions.

Open-source, self-hostable, and built for teams that need fast, trusted answers. Connect any database. Extend through adapters.

MIT LicensedAdapter-firstSelf-hostable