When AI does your data work, the work needs somewhere to live.
Answers in a chat evaporate — the cleaned data, the queries, the reasoning, gone with the conversation, and the numbers come without receipts. In Data Aqmen, an agent works in a real workspace: data lands clean and cited, every conclusion keeps its evidence, and each session builds on the last.
The loop
Four steps, each one leaving a trail the next can build on.
Import cited data
Drop CSV, JSON, or Parquet files — or let an agent research and load the data for you. Every dataset carries sources, so each number stays traceable to where it came from.
Transform with SQL
Materialize clean tables from raw datasets with SQL transformations. Each transformation and its output are one artifact: the result, the schema, and the SQL that produced it.
Query & build views
Save SELECTs with charts that stay current — and describe the screen you want: the agent builds views, focused interactive dashboards over your live data, one per question.
Record insights
Pin short conclusions to the query or dataset that demonstrates them. When the underlying data changes, the insight is flagged stale so you know to re-check it.
Agent-first, by design
Data Aqmen ships a built-in MCP server, so any AI model connects directly to your workspace — no glue code. An agent reads what's already there, loads researched data with citations, files its work into collections, and records conclusions with receipts: every claim linked to the query or dataset that demonstrates it.
- Reads the workspace: datasets, schemas, lineage, and prior insights before it starts.
- Loads cited data: researched tables arrive with sources attached, not as bare numbers.
- Builds real artifacts: transformations, saved queries, charts, and views you keep.
- Records insights with evidence — flagged stale when the data underneath moves.
Load the latest EU HR-tech funding data and tell me where the money is going.
Done — here's what I set up in your workspace:
- Created dataset eu_hrtech_funding — 214 rows, 4 sources cited
- Built transformation funding_by_segment
- Saved query “Funding share by segment, 2021–2025” with a bar chart
- Built view “Funding overview” — an interactive dashboard on your live data
- Recorded 3 insights, pinned to their evidence — ready for your review
Core HR captures the largest funding share. Everything is cited and in the Activity feed.
Everything keeps its receipts
Data, the SQL that shapes it, the charts that show it, and the conclusions drawn from it — all connected, all auditable.
Datasets with provenance
Every imported table carries citations for where the data came from and what judgement was applied.
Transformations + lineage
SQL that materializes new tables, with a lineage graph tracing every output back to its sources.
Views: dashboards on demand
Describe the screen you want and an agent builds it — a focused, interactive view over live workspace data, one per question. Saved queries stay the citable chart layer underneath.
Insights with staleness
Conclusions pinned to their evidence — flagged automatically when the data underneath them changes.
Collections
Folders that keep a growing workspace organized, for humans and for agents scoping their work.
Activity log
An append-only record of who did what — every import, transformation, and insight, agent work included.
Questions, answered
The short version of what people ask before they start.
Why not just chat with my files?
You can — once. The agent re-parses everything each session, the cleaned data and reasoning vanish with the chat, and there is no way to trace how a number was produced. In a workspace, the same ask leaves durable artifacts: cited datasets, inspectable SQL, charts that re-run, conclusions pinned to their evidence. The next question starts from all of it.
Where does my data live?
In managed workspaces on Data Aqmen's infrastructure. Files you import become versioned tables in storage that is isolated per workspace, and every import keeps its history — you can always see what a dataset looked like before.
Do I need to know SQL?
It helps, but it isn't required. The AI agent writes SQL for you, and the UI shows results as tables and charts. Everything the agent builds — transformations, saved queries, views — stays inspectable and editable, so you can learn from it or take over at any point.
Which AI models work with it?
Any MCP-capable client: Claude, ChatGPT, and others. Connecting uses OAuth sign-in — no API keys to create or paste. The setup steps live in the MCP integration guide.
Can the AI change my data without me knowing?
No. Agents act as your signed-in user, and every action they take lands in the Activity feed. Queries are read-only by design, datasets are versioned so nothing is silently overwritten, and insights an agent records wait for your validation before they count as reviewed.
What file formats can I import?
CSV, TSV, JSON, NDJSON, and Parquet — gzip-compressed files are fine too. Each import becomes a queryable table with its schema detected for you.
How do I try it?
Sign in and create a workspace with demo data: a complete worked market-research example with every row cited. Explore it, query it, connect an agent to it — and delete it whenever you're done.
Start with the demo workspace.
A complete worked example — real market research built from public sources, every row cited. See how the pieces fit in five minutes, then connect an AI agent and let it work.