Columns AI is a data storytelling platform that helps you move from raw data to clear, shareable insights in minutes. The docs split the product into Columns Drive, Columns Flow, and Columns Graph. Drive turns uploaded files into a dataset. Flow is a visual, node-based workspace for building data workflows. Graph is where the docs say you annotate, style, and share. The April 2, 2026 post says the platform is live at columns.ai and says to try it for free today.
Use Columns when the work is a dataset you already have and a story you still need to draw. The introduction lists Google Sheets, Excel, Notion, Airtable, CSV and TSV uploads, Snowflake, Databricks, BigQuery, Postgres, and HTTP endpoints as source types. Flow then asks you to describe a transformation, generate a compute plan, preview a sample, and approve it before downstream nodes change. That is a different job from a chat that only answers a question about a spreadsheet. The public docs and the launch post do not show a paid price. They do say to try it for free today at https://columns.ai.
A good fit for
Analysts who want joins, lookups, and aggregates written as plain-language intent.
Teams uploading receipts, invoices, or screenshots into a Drive folder for one dataset.
People publishing a chart with annotations, a link, an embed, or a Slack share.
Know the trade-offs
Connected sources stay in your systems and sync on demand. Uploaded CSV and TSV files are stored in Columns cloud storage.
Every Drive file has a public link by default until you turn that link off.
MySQL and BigQuery are listed as upcoming examples, not as finished connectors on the overview page.
Drive files, Flow nodes, and Graph stories.
Connect a source, then keep uploads and live data apart
The introduction says Columns combines data connection, transformation, visualization, and communication in one product, for analysts, operators, founders, and teams that need to make decisions quickly. Connect Data lists data apps such as Google Sheets, Excel, Notion, and Airtable, file uploads of CSV and TSV, SQL sources such as Snowflake, Databricks, BigQuery, Postgres, and others, and HTTP endpoints that return JSON or a file. The data overview narrows the storage rule. Upload sources (CSV/TSV) are stored in Columns cloud storage. Connected sources (such as Sheets, Notion, SQL, APIs) remain in your own systems and are synced on demand. Current connectors named on that overview are CSV/TSV uploads, Columns Drive, Google Sheets, HTTP endpoints, Notion databases, Airtable, Postgres, and Snowflake, with additional connectors in progress. Upcoming examples are MySQL and BigQuery. Chart types in the docs navigation include map, timeline, bar, line, pie, scatter, tree, bar race, radar, boxplot, and word cloud. Those names are the menu, not a guarantee that every source can fill every chart.
Build a Flow from nodes, edges, and an approved plan
Columns Flow is a visual, node-based workspace for building data workflows in Columns. Each node represents a dataset, and each edge represents a transformation between nodes. You can sync source data, preview outputs, and save visuals from any node. The documented path starts at an input node. Select a data source, sync so schema and samples load, add a node, connect it, and open the edge. Describe your transformation intent and click Generate to produce a compute plan. Preview the plan and sample output, then click Approve to update the target node. Repeat for multi-step workflows. A built-in starter example is included. Multi-source operations are named as Union, Join, Lookup, and Except. Join is two or more datasets on join keys and is called equivalent to a SQL inner join. Lookup keeps all rows from the primary source and is called equivalent to a left join. Except keeps primary rows with no match and is called equivalent to an anti join. Intent examples include removing a header row, converting a date to a month, filling empty sales with 0, keeping only date, region, and sales, totaling sales by region on a monthly timeline, and leaving the top 10 results for each month. Tips say to use column names as they appear in the schema, to state the output shape, and to put filters, limits, and sorting in the same sentence.
Columns Drive works like a cloud drive, similar to Google Drive, OneDrive, or Dropbox. The difference is that it automatically extracts structured data from uploaded files so you can use them immediately. The file types written next to that sentence are.png,.jpg,.pdf, and.txt. The Drive page says it is not general-purpose storage. It is built for data extraction, combining OCR and modern AI so unstructured files become analysis-ready datasets. Product screenshots, receipts, invoices, bank statements, and similar files go in a folder, and extracted data can be downloaded later. Every uploaded file has a unique public link by default, but you can turn it off if you want. In Flow, a Drive folder can be the input. The folder becomes an auto-generated dataset, and Columns merges extraction outputs from all files in the folder into one unified dataset. Extraction is context-aware: folder name and previously extracted files help improve schema consistency for new uploads. The launch post adds Excel sheets to the upload list and says the extractions for a folder are compiled into one CSV you can download or send into Flow. The post's user story about hundreds of vendor files is a vendor anecdote.
The introduction says Columns is also built for presentation-quality storytelling. You add annotations, images, shapes, arrows, and widgets, highlight points with color, contrast, and layout, and fine-tune typography, spacing, and chart settings. A tip says the graph canvas is SVG-based, so you can click and style almost any element directly. Another tip says Columns uses your context and interaction history to provide more relevant recommendations over time. Analyze and Visualize lists automatic chart suggestions, queries with dimensions, metrics, and filters, and visual controls on the canvas. Share and Collaborate lists four exits: publish with a link, embed in external pages, share to Slack, and export assets for reports and presentations. The launch post names Columns Graph as the visualizer and says you can bring your own brand assets, custom colors, and typography so published dashboards follow that brand. Flow's plain-language section on the same post says you can set up cleaning, transformation, or analysis intent, including aggregations and unions and joins, then automate the flow and subscribe to alerts for special events or anomalies. That alert line is the post's description of cruise control, not a price.
Follows the Flow intent examples rather than a blank canvas guess. Connect a sheet or a CSV, sync the input node, and write an intent that uses the real column names. The docs' own sample is to get total sales by region along a monthly timeline, then sort by total sales and leave only the top 10 results for each month. Generate the compute plan, preview the sample output, and approve the edge before you trust the next node. Visualize the node once it has schema and sample data. If the date field is a long timestamp, the same page shows a separate example that converts it to a month value in yyyy-mm form, and another that fills 0 when sales is empty. Put the filter, the limit, and the sort in one sentence when you can.
Drop invoices into a Drive folder
Another suggested workflow is for files that are not already rows. Create a Drive folder whose name describes the documents, upload png, jpg, pdf, or txt files, and download the extracted spreadsheet when you need it. The Drive page says the folder can also be selected inside Flow, where extracted files are merged into one dataset for later nodes. Folder name and earlier files are supposed to steady the schema for the next upload. The public-link default matters before you share a folder. Every uploaded file has a unique public link by default, but you can turn it off. The launch post's story about a person dropping more than 100 vendor files is the company's anecdote. Treat it as a suggested illustration of the folder model, not as a measured extraction rate.
Publish the chart after the pipeline is approved
A suggested last step is the story, not another transform. After Approve has updated the node, open the canvas, accept or adjust a chart suggestion, and add the annotation, arrow, or widget the introduction lists. The SVG note means the docs expect you to click elements and change type, spacing, and color there. Bring brand colors and type if you are in Graph, which the launch post describes as the personalized visualizer. Then pick a share format the introduction actually names: a link, an embed, a Slack share, or an export for a report. The pages do not describe a scheduled email digest beyond the Flow post's alert subscription for anomalies.
How to get started with Columns
1
Choose the source
Upload a CSV or TSV, or connect Sheets, Notion, Postgres, Snowflake, Drive, or an HTTP endpoint.
2
Sync the input node
In Flow, sync so the node has schema and sample rows before you add an edge.
3
Generate and approve
Describe the intent, click Generate, preview the sample, then click Approve.
4
Style and share
Annotate the SVG canvas, then publish a link, an embed, a Slack share, or an export.
You want a predictive model on a table more than a narrated chart.
Columns docs center on connecting data, approving a Flow transform, and designing a story. Choose Akkio when the job is a model rather than Drive extraction plus a Graph canvas.
Columns: common questions
What are the three products on the Columns pages?
The docs and the April 2, 2026 post name Columns Drive, Columns Flow, and Columns Graph. Drive extracts files, Flow transforms nodes, and Graph is the visualizer for the story you share.
Upload sources (CSV/TSV) are stored in Columns cloud storage. Connected sources such as Sheets, Notion, SQL, and HTTP endpoints remain in your own systems and are synced on demand.
You describe the transformation, click Generate to produce a compute plan, preview the plan and sample output, then click Approve to update the target node. Downstream nodes update from that approval, not from the typed sentence alone.
The Flow page names Union for compatible schemas, Join as an inner join on keys, Lookup as a left join, and Except as an anti join. It also shows aggregate, pivot, filter, and top-10 examples in the intent list.
NavHunt did not run Columns. No file was uploaded, no Flow was approved, and no chart was published. This profile quotes the docs introduction, the Flow overview, the data overview, the Drive overview, and the April 2, 2026 product-line post public for this check. Vendor stories on that post are not treated as measured timings. This description is based on Columns's official materials reviewed on 2026-10-06, including Columns docs introduction, Columns Flow, Columns data overview, Columns Drive.