▚ Table Intelligence

Data analysis agent

The Data analysis agent brings real data science into your AI agent — cohorts, RFM, retention, forecasting, classification, association mining and causal inference — and runs it on your machine. You ask questions in plain language; it does the statistics and explains the answer, including how much to trust it.

It’s one of the agents in the Table Intelligence suite. Everything here runs locally; your data never leaves your computer (see Privacy & data).

What you can ask

Once the server is installed, just point your agent at a CSV — orders, transactions, donations, usage, anything tabular — and ask. Some examples, by what you’re trying to do:

Understand a table

“Analyze orders.csv with Table Intelligence.” “Profile this file — what’s in each column, what’s missing, what’s unusual?”

Segment your customers

“Run an RFM segmentation on customer_id, order_date, amount.” “Show me monthly retention cohorts. Where are we leaking customers?”

Find what drives a number

“What’s most associated with a high order value?” “Did the price change cause the drop in orders?”

Predict

“Which customers are likely to churn?” “Forecast next quarter’s revenue, and tell me how confident that is.”

Explain a result

“Why did the model flag this customer as high-risk?”

Each answer comes back with a plain-language summary and the numbers behind it.

Every result is honest about its own limits

This is the bit that separates a real analysis from a confident-sounding guess. Every answer carries a trust level and its caveats:

"trust": { "level": "moderate",
           "caveats": ["Moderate sample (120 rows) — reasonable but not definitive.",
                       "This is a before/after comparison, not a causal test."] }

And when the data can’t actually support the question — too few rows, no variation, a failed robustness check — the agent declines instead of inventing a number:

"method": "causal_effect", "declined": true,
"trust": { "level": "none",
           "decline_reason": "Only 18 rows — too few for a credible causal estimate." }

Take a decline seriously: it’s the honest answer, and it’s more useful than a confident number that isn’t real. Trust levels range from high → moderate → low → none.

Where the output goes

By default the answers just appear in your chat. When you want to keep something, ask the agent to save it as a report — it lands on your dashboard, where you can organise it into folders and share a public link.

Next

Table Intelligence — built by Shubham Randive. Questions? Get in touch.