How to use DataUnmess

DataUnmess is an AI artifacts platform. Unlike AI chatbots that produce text, DataUnmess produces permanent, shareable visual assets — dashboards, flowcharts, data explorations, and lineage graphs — from natural language.

Connect your data (CSV, Excel, Google Sheets, PostgreSQL, MySQL, REST APIs, MCP servers) and describe what you want. The AI builds it as an interactive artifact you can save, reload, and share. No AI tokens consumed on reload.

1. 1

   Open the Data tab

   It's on the right-hand panel.

2. 2

   Import your file

   Click + Import and drop a CSV or Excel file, or paste a Google Sheets URL.

3. 3

   Ask a question

   Switch to the Chat tab and ask, e.g. "show me sales by category as a bar chart".

4. 4

   Refine by chatting

   Continue the conversation to add, modify, or remove cards. The AI sees the current dashboard state.

5. 5

   Save

   Click Save in the toolbar and pick a folder. Saved dashboards reload exactly as you left them.

DataUnmess supports multiple data sources. Open the Data tab in the right panel to manage imports and connections.

File imports

• CSV — drag and drop or click to browse. Parsed instantly in the browser.
• Excel (.xlsx, .xls) — sheet selection, header row picker, and column mapping preview.
• Google Sheets (import) — sign in with Google, paste a sheet URL to import a one-shot CSV snapshot. Works with any sheet in your Drive.

For live data that re-queries on every prompt, use the Google Sheets connection (under Add Connection → SaaS) instead of the import — see APIs & protocols below.

Database connections

• PostgreSQL — Neon, Supabase, or any Postgres-compatible database. Schema discovery shows tables and columns.
• MySQL — MariaDB, PlanetScale, TiDB. Same schema discovery and SQL query support.

The AI writes SQL queries and charts the results. All credentials are encrypted with AES-256-GCM.

APIs & protocols

• REST API — any JSON endpoint. Configure base URL and auth (Bearer, API key, Basic).
• MCP Servers — Model Context Protocol for AI-powered external tools and data sources.
• GitHub — connect repositories to analyze code, technologies, and architecture.
• Google Sheets (live) — OAuth-authorized connection that lets the AI list spreadsheets in your Drive and read ranges or whole sheets on every request. Refresh tokens are stored encrypted; access tokens rotate automatically.

Coming soon

Snowflake, BigQuery, Stripe, and HubSpot connectors are in development. They appear in the connector catalog with a "Coming Soon" badge.

Cards

• Drag — hover a card, grab the handle in its top-left corner, and drop it on any grid slot.
• Resize — hover to reveal bottom, right, and corner handles; drag to snap to new widths/heights.
• Settings — the gear icon opens inline config (value format, grid lines, sort order, hide title, etc). Changes apply live — roll back with Ctrl+Z.
• Delete — the × in the top-right.
• Copy title — click a card title.

Toolbar

• Save — opens the save dialog, pre-filled with the current name and folder.
• Delete — permanently removes the currently-loaded dashboard (with confirmation).
• Align — resets every card back to its default size/position.
• Undo / Redo — Ctrl+Z and Ctrl+Shift+Z.

Right panel

• Chat — ask anything about your data.
• Recent — browse and restore past sessions.
• Data — imported datasets and a preview of the first 10 rows (horizontally scrollable).
• Style — pick a theme or create a custom one based on any built-in theme.

DataUnmess is AI-first: everything you can do in the UI, you can also ask the AI.

Creating charts

Modifying existing cards

Changing a chart's type or adding a derived series

Reference cards by the on-card id: N badge — same number on chat and MCP. Type changes (e.g. bar → combo) are composed as remove_chart + build_chart with the original layout preserved. Derived series (deltas, running totals) are computed in SQL via query_data first, then rendered as a combo (bars for the original series, line for the derived one).

Repointing a dashboard at a new dataset

After a transformation produces a cleaned v2 dataset, redirect every chart on a dashboard to it in one shot — no need to rebuild the layout. The AI calls swap_dashboard_dataset, drops cached chart data so the next reload re-queries the new source, and warns you about any charts referencing columns the new dataset doesn't have.

Themes

Loading data inline

Working across tabs

The studio has four tabs — the AI tailors its output per tab:

Unlike traditional AI chat, DataUnmess produces visual artifacts — not text descriptions. Every response generates objects that render directly on the dashboard canvas.

Artifact types

• Charts — 20+ types (bar, line, donut, area, scatter, treemap, funnel, radar, sankey, heatmap, waterfall, gantt, ridgeline, streamgraph, data-table, and more)
• KPI cards — single-metric highlights with trend indicators and % change
• Analysis panels — written insights with highlight cards showing key metrics
• Dashboards— complete multi-chart layouts from a single prompt ("build me an overview")

Data science exploration

Use conversation to explore your data like a notebook. The AI aggregates, filters, groups, and joins — you see the results as charts instantly. Ask follow-up questions to drill down, pivot, or compare.

Lineage & flows

The Lineage view traces how datasets flow into charts and dashboards as an interactive node graph. The Flow Editor lets you build visual pipelines and transformation diagrams with 9 node shapes (rect, diamond, ellipse, hexagon, parallelogram, cylinder, cloud, document, text) and lucide icons per node. Accepts either structured JSON or a Mermaid flowchart TD|LR block and exports back to Mermaid.

Executable Data Pipelines

Beyond diagrams, the Data Pipelines tab runs real data-cleaning pipelines across Google Sheets, files, REST APIs, and warehouses. Ask the AI to clean a sheet — "dedupe by email", "split full_name into first / last", "normalize phone numbers to E.164" — or to import a SaaS API ("pull my Stripe charges from the last 30 days"). The engine is chosen automatically: Python (pandas) by default for files, sheets, APIs, and cross-connection flows; SQL pushdown when source and sink are the same database (rows never leave the warehouse, compiled from filter / select / derive / aggregate DSL ops into a single CTE chain); and DuckDB only when you explicitly pin it for huge local files. The AI calls validate_transform_flow first to show you a sample, then run_transform_flow to materialize a new destination — a new spreadsheet in Drive (the app uses the narrow drive.file OAuth scope, which only allows write access to files the app created), a new dataset, or a new table in your DB. Dataset sinks can land rows in your local CSV store (the default for local dev) or in managed Postgres — your workspace gets a dedicated ws_<id> schema on Neon, so customers without their own database can still persist the output. Every run is recorded with per-step row counts and, on failure, the compiled Python source or SQL of the failing step so you can ask the AI to patch and re-run. Pipelines can be renamed, moved between folders, or permanently deleted with update_transformation and delete_pipeline — deletion requires explicit confirmation (the MCP server pops a Yes/No prompt; non-interactive clients pass a typed DELETE <id> token), and cascades the schedule + run history.

Ask "how much of my plan have I used this month?" at any time — the AI calls get_workspace_usageand answers with the workspace's plan, current row count, bytes stored, and sync runs against the plan's quotas. Useful before kicking off a big import, or right after a run is refused by a quota gate.

Persistence

Artifacts persist across sessions and survive page reloads with zero AI tokens consumed. Sessions restore from the database. Saved dashboards re-execute local queries against the original data — no AI call needed.

Connect DataUnmess to Codex Desktop, Claude Desktop, Claude Code, ChatGPT, Gemini, Cursor, or Windsurfby pasting our hosted MCP endpoint + an MCP key into your client's config. The AI then calls DataUnmess tools directly — building dashboards, running queries, generating analysis — with your data.

No install, no clone, no code on your machine. All DataUnmess logic stays on our servers.

{
  "mcpServers": {
    "dataunmess": {
      "url": "https://app.dataunmess.ai/api/mcp",
      "headers": {
        "Authorization": "Bearer <YOUR_MCP_KEY>"
      }
    }
  }
}

Full connect guide →

DataUnmess Memory is the workspace memory layer used by your MCP-connected AI. It stores company context, KPI memory, business glossary notes, and data catalog knowledge as agent-readable markdown. The goal is simple: every useful piece of knowledge you add should make future answers, research, and artifact creation smarter.

How it starts

During MCP onboarding, ask your AI to start DataUnmess Memory onboarding. The AI should ask for your company website, research public pages, and seed the workspace memory with company context, products, glossary terms, standard KPIs, and useful business language.

Where it is used

• MCP answers and research - when you ask business questions, the AI can search memory before guessing from raw data.
• Dashboard creation - dashboards can reuse known KPI names, definitions, formulas, business context, and source-of-truth cards.
• User questions- questions like "what is my Patrimonio Liquido?" can resolve from KPI memory and snapshots instead of rediscovering dashboards.
• Flowcharts - process diagrams can reuse business terms, company products, teams, systems, and operating rules already captured in memory.
• Data pipelines - transformation and pipeline work can add or reuse catalog knowledge such as dataset grain, key columns, trusted use cases, and known data caveats.

How knowledge grows

Knowledge can come from onboarding, manual notes, approved conflicts, dashboards, flowcharts, datasets, and pipelines. New low-risk information is learned automatically. Review is only needed when the AI finds conflicting definitions, competing sources of truth, missing fields, or uncertain data meaning.

Markdown-first storage

DataUnmess Memory is designed to be readable by agents and by humans. KPI memory, KPI snapshots, company knowledge, catalog entries, and review items are written as markdown files under the workspace knowledge folder, with lightweight indexes only for speed.

DataUnmess ships with six built-in themes: Carbon, Candy, Neon, Ocean, Sunset, Emerald. Built-in themes are read-only — you can't edit them directly.

To make your own: open the Style tab → click + New→ give it a name → pick a base theme. The new theme appears under "Custom" with an Edit button that lets you tweak every color. Custom themes are saved in your browser.

The left sidebar lists saved dashboards grouped by folder. Folders can be nested — create a subfolder by hovering a folder and clicking the + that appears.

• Move a dashboard — drag it onto any folder.
• Delete a dashboard — hover and click the ×.
• Delete an empty folder — hover and click the ×.
• Resize sidebar — drag its right edge.