binggioTM

Enterprise Context Intelligence

Answers from your data you can trust.

Binggio answers questions using only the meanings a person has approved: which column is the key, how tables link, what "delinquent" means. When it can't be sure, it refuses and tells you why.

$ pip install binggio
Watch the 2-minute demo

Free and open source · Runs on your computer · Python 3.10+

What is the total outstanding balance of delinquent loans for each product?

AUTO_LOAN98,730,000
BUSINESS_LOAN123,000,000
HOME_LOAN105,595,000
PERSONAL_LOAN89,825,000
Fromloan_accounts · loan_repayments TermDelinquent loan: at least one overdue repayment instalment on the as-of date Approved As of2026-10-19 StepsStarted from loan_accounts; applied "Delinquent loan"; summed outstanding_balance; grouped by loan_product

A real answer from Binggio's sample bank. Every answer carries its proof.

The problem

AI on company data guesses. Silently.

Real data has several columns that could mean the same thing. Text-to-SQL picks one, runs it, and returns a confident number. Nobody approved that choice, and nobody can see it was made.

Binggio puts people back in charge of meaning. It suggests, you approve, and answers use nothing else.

Which one is "delinquent"?

loan_delinquency.days_past_due
20 · 45 · 90 …
loan_delinquency.delinquency_band
DPD_1_30 · DPD_31_60 · DPD_61_PLUS
loan_repayments.payment_status
PAID · MISSED · PARTIAL · DUE

How it works

From raw tables to governed answers in four steps.

1

Connect

Point Binggio at CSV files, an Excel workbook or a database. It copies the data read-only into a workspace on your computer.

2

Discover

It profiles every column and suggests keys, links between tables and what each column means, with the evidence for each.

3

Approve

You review and approve: column roles, relationships and business terms like "delinquent loan". Every decision records who and why.

4

Ask

Ask in plain words or from an AI agent. Answers use only what was approved and show their proof, or refuse with the reason.

What you get

Built for people who have to stand behind the number.

Proof with every answer

The tables used, the exact approved definition, the as-of date and every step, plus the query that ran.

Refusals, not guesses

If a question needs a meaning nobody approved, Binggio says what is missing instead of picking something.

Relationships found for you

Keys and links across tables and systems, suggested with match evidence. Weak matches are left for you to decide.

Approvals that expire

Approvals are bound to the data they were made on. When the data changes, they go stale until someone confirms them again.

The data story

Statistical tests surface what drives your totals, where they concentrate and how they change, from approved measures only.

Follow-up questions

Ask "which of these…" and Binggio works on the previous answer, keeping the same governed meanings.

For AI agents

Don't let your agent write SQL.

Binggio is an MCP server. Claude Desktop, Claude Code, Cursor and VS Code see only your approved vocabulary. The agent builds a plan; Binggio checks it and runs it read-only, or refuses with the reason.

No SQL accepted, nothing written to your sources
Invented columns are refused, never substituted
Agents can propose a meaning; only a person can approve it

Your agent receives the answers it asks for, so it can show them to you. Binggio itself never sends rows to a model.

terminal
$ binggio connect claude-desktop -w ~/my-bank
Added Binggio in claude_desktop_config.json

// agent asks for a column nobody approved
{"status": "refused",
 "message": "Unapproved field: CHURN_RISK",
 "guidance": "Not answered. Tell the user the reason;
   do not guess or substitute another column."}

Security by design

Your data stays where it is.

Runs on your machine

The app runs locally on 127.0.0.1. No cloud service, no account.

Read-only on sources

Nothing is ever written back to your files or databases.

Secrets never stored

Database passwords and model keys stay in memory only, never on disk.

A full audit trail

Every approval, rejection and revocation is recorded with a name and a note.

Connects to CSV files, Excel workbooks, SQLite, SQL Server, Oracle and PostgreSQL.

What the AI model sees, and what it never sees

Binggio uses a language model for one step only: turning a question into a structured plan. Checking the plan, running it read-only on your data and building the answer all happen inside Binggio, in your environment.

Sent to the model

The question as typed

Names, roles and types of approved tables and columns

Approved links and business-term definitions

Category labels of approved columns, such as PREMIUM or MISSED

Never sent to the model

Rows from any table

Answers and result values

Database passwords and connection details

Anything from tables or columns nobody approved

Run it with no model at all, using saved plans or an AI agent over MCP. Choosing your own model provider, including one in your own cloud or on your own servers, is coming soon.

Open source, with a path for business.

Community

Free

The full product under the GNU AGPL v3. Use it, study it, change it, and share your improvements.

Get it on GitHub

Commercial

Talk to us

A commercial licence for teams that want to embed Binggio in a closed product or service, plus help rolling it out on your data.

ashish@binggio.com

See it on your own data in five minutes.

$ pip install binggio
$ binggio ui

Opens in your browser with a ready-made sample bank, or start with your own files.