twoHelixestwoHelixes Start free
Your data. A new perspective.

Turn connected data into clear decisions.

Ask in plain language. twoHelixes finds the right tables, joins and shapes them in real Python, and draws a chart that does not mislead — streaming its reasoning while it works, so every step is yours to check.

For operators, analysts and founders who have the data but not a BI team — and for data teams tired of being the queue.

Ask one question now without an account. Sign in for 10 free charts a month — no card, and the sample data is already loaded.

Start where your data lives CSV & ExcelPostgreSQLSnowflakeBigQueryHTTP APIs Explore connections ↗

Less digging. More discovering.

A question in.
A new perspective out.

Follow the data from question to finding. Inspect the steps, adjust the chart, and keep asking.

twoHelixes / Revenue explorationSample analysis
↳How did revenue trend by region?
East overtook South in July0200400600800JanMarMayJulSepNovMonthNorthSouthEast

See the story behind the lines. In this illustrative dataset, East moves ahead of South in July. Rendered by the same chart engine used in the app.

How it is different

Not a chatbot over one table.

Most AI analytics answers from a single pre-modelled table. twoHelixes does the analyst’s part too: finding, joining and shaping.

Joined and shaped, not just queried

It picks join keys by measured value overlap, checks the row count survived, and reshapes in pandas — resampling at the data’s own grain.

Reasoning you can watch

Every stage streams as it runs: what it chose, the code it executed, the chart forms it rejected. A failure stays open on the step that failed.

Charts audited by rule

Fixed colour order, no dual axes, colour-blind-safe series caps. Anything approximated is printed beside the chart, never hidden.

Edits that cost nothing

Chart type, axes, grouping, sort, top N, log scale, labels and number format change instantly with no model call. Only a new question spends one.

  1. Finding the datawhich tables can answer this, and how sure it is
  2. Joiningthe key it chose, the value overlap it measured, rows kept
  3. Shaping the datathe pandas it ran, in full, editable
  4. Choosing the chartthe form it picked and the forms it rejected
  5. Applying defaultspalette, labels and a re-audit of the finished figure
  6. A chart titled with the findingtrace folds to one line; SQL, Python and notebook one click away

One connected workspace

Go beyond the chart.

From a quick business question to hands-on analysis, work at the depth your decision needs.

Explore & explain

Ask in plain language. Compare performance, uncover trends, and inspect the transformations behind each result.

Explore a dataset ↗

Build a shared view

Bring charts into dashboards. Share a link with your team or export a figure for your next presentation.

Discover dashboards ↗

Go deeper with code

Work in SQL, sheets, and Python notebooks. Use agents for multi-step analysis, with editable code and reusable outputs.

Explore the toolkit ↗

SQL workbench

Schema-aware completions that never wait on a model, generated SQL you can edit and rerun for free, read-only execution and full history.

Open the editor ↗

Sheets with an agent

A spreadsheet whose agent proposes cell edits and charts. You preview every change and approve it; model output never overwrites a workbook.

Try Sheets ↗

Machines when you need them

CPU sessions are included in paid plans. GPUs from $0.60 an hour, billed by the minute and stopped when idle — for forecasting, embeddings and heavy notebooks.

See machine rates ↗
Frontier reasoning. Inspectable results.

deepseek-v4-flash powers the configured analysis route, with task-specific routing and fallback models. Chart validation stays in code, independent of the model.

Chart quality

Charts that do not mislead

The rules are enforced in code, not left to taste. Every chart is audited against them before you see it — including the three below.

Your data

Your sources. One place to make sense of them.

Drag in a CSV, Excel file or Parquet extract and ask a question thirty seconds later. Or connect the warehouse and leave the rows where they are — everything is read-only, and generated SQL is checked before it ever reaches a driver.

PostgreSQLMySQLSQL ServerSnowflakeBigQueryRedshiftClickHouseTrinoOracleDuckDBSQLiteMongoDBElasticsearchHTTP APIsCSVTSVJSONParquetExcelOpenDocument.gz.zip
orders.csv uploaded · 41k rows
warehouse · Snowflake read-only
regions joined on region_id, 98% overlap
SELECT … checked read-only

After the first chart

The chart is a starting point, not a screenshot

Change it for free

Swap the axes, the chart type, the grouping or the aggregation by hand and it redraws — no model call, no credit spent. Or say what you want changed in words and let the edit stage apply it.

Real Python underneath

The shaping runs in a sandboxed interpreter with pandas, polars, numpy, scipy, scikit-learn and statsmodels — so “fit a trend and forecast the next quarter” is a question, not a feature request. The code is shown and editable.

A real SQL editor

For when you already know the query. Schema-aware completions answer instantly without waiting on a model, plus generated SQL, per-source suggested questions and saved history.

Take it with you

SVG, PNG or CSV; pin it to a dashboard with a share link that needs no account; export the whole run as a Jupyter .ipynb or a marimo notebook — both run here as well as on your laptop; or call the same pipeline over HTTP.

An honest comparison

Pick by who does the work.

Looker, ThoughtSpot and Hex are good products built for different people. Here is where each one wins.

twoHelixesLookerThoughtSpotHex
Built forAnyone with a question and a file or databaseData teams governing metrics in LookMLBusiness users searching modelled warehouse dataAnalysts writing SQL and Python together
Before the first chartDrop a CSV or connect read-only; minutesA LookML model built by the data teamData modelled into ThoughtSpot firstWrite the query yourself
Plain-language questionsYes, including joins and reshaping across sourcesOver modelled ExploresOver modelled data, its core strengthAI assist inside notebook cells
Shows its workStreamed reasoning, the Python it ran, rejected chart formsThe semantic model is the source of truthSearch tokens behind each answerCode cells are the work
PricingPublic and self-serve, free tier, no seat minimumSales-ledPublished entry tiers, enterprise via salesFree tier plus published team plans
Where it winsFastest path from a messy question to a joined, honest chartOne governed definition of every metric at enterprise scaleSearch at scale with fine-grained security on a warehouseCollaborative notebooks and data apps for a data team

From each vendor’s public documentation, October 2026. If you need a governed semantic layer for thousands of users, buy Looker; if you want an answer from your own data this afternoon, start here.

Ask which channel brings the most signups?
Organic drives 38% of signups01k2k3k4k5kOrganicEmailSocialChannel

Get started

Your next decision starts with a question.

Ask one question now without an account. Sign in for 10 free charts a month — no card, and the sample data is already loaded.

Explore sample data ↗ See plans

Frequently asked questions

What is twoHelixes?

An AI data analyst. Ask a question in plain language and it finds the data, joins it, shapes it, picks a chart and streams its reasoning while it works.

Do I need an account to try it?

No. Ask one question on sample data without signing up. Free accounts get 10 AI charts a month with no card.

Which data sources does it connect to?

CSV, TSV, JSON, Excel, OpenDocument and Parquet files, plus PostgreSQL, MySQL, SQL Server, Oracle, Snowflake, BigQuery, Redshift, ClickHouse, Trino, DuckDB, SQLite, MongoDB, Elasticsearch and HTTP APIs. All connections are read-only.

Can it replace Google Analytics, Mixpanel or Amplitude?

For question-driven product analytics, yes: the first-party collector accepts GA4, Segment, Mixpanel and Amplitude shaped events. Audiences and ad exports are not supported; the parity map lists every gap.

Is there an API for agents?

Yes. Every chart is one HTTP call. See /openapi.json, /llms.txt and /docs.

How is it priced?

Free, then Plus at $19, Pro at $49 and Team at $149 a month, with pay-as-you-go credits (1 credit is 1 US cent) that never cut you off mid-month.