Simple questions, answers spread across many screens

Answers are spread across many screens and tables.

Asking someone, or opening five pages, takes too long for simple questions like "how did last week's posts do?"

We wanted to ask in plain English and get a short, correct answer, without giving an AI the power to change anything.

A chat that can look, but never touch

Ask AI is a chat inside our own apps. It was built for our own companies, BinaryBits and TheDataHQ, and it is used every day.

The product screen, shown with test data. Real screens use the same layout.
💬
A chat inside the app
Plain questions, plain answers.
🔒
Read-only access
The agent can look up data but can never change anything.
🧭
Step-by-step lookup
The agent finds the right data, then explains the answer.
🔁
Conversations that continue
Pick up where you left off.
⚖️
Model per feature
The owner chooses which Claude model each feature uses, balancing quality and cost.

From question to grounded answer

AI agentAutomated stepPersonData or serviceQuestionin plain EnglishAgentplans the lookupRead-only data accessnothing can be changedAgentchecks and explainsAnswershort and groundedquestionlookupsdatachecked answer
Read only: the agent can look things up, but nothing can be changed.
1
Ask
Type a question in plain English inside the app.
2
The agent plans what to look up
It works out which data answers the question before reading anything.
3
It reads the data
Through read-only access, so nothing can be changed.
4
It explains the answer
A short answer, grounded in what it looked up. The conversation can continue from there.

Safe by design, grounded in data

Safety comes from the design, not from instructions: the agent simply cannot change data. The agents run on Anthropic's Claude, through Claude Code skills.

WhoWhat it does
Planner (AI)Works out what to look up to answer the question.
Read-only access (automated)Returns the data the agent asks for. It cannot change anything.
Explainer (AI)Checks what was found and writes a short answer grounded in it.
The owner (person)Asks the questions and chooses which Claude model each feature uses.
Safety

Read-only access by design. There is no possibility of changes.

Trustworthy answers

Short answers grounded in what was looked up, so they don't wander.

Cost

The owner chooses the right model per feature, balancing quality and cost.

Continuity

Conversations continue where they left off.

Technologies used

Our own Python code, a relational SQL database, and Anthropic's Claude.

Python Claude Relational SQL database

Challenges and how we solved them

❌ Challenge
Useful access without risk.
✓ What we did
Read-only access by design.
Result: No possibility of changes.
❌ Challenge
Answers that wander.
✓ What we did
Short answers grounded in what was looked up.
Result: Trustworthy replies.

Answers without opening five screens

Daily
In daily use
Read-only
No possibility of changes
Per feature
Model chosen to balance quality and cost

Ask AI is in daily use. Business results: not applicable for this internal tool.

Last updated: November 2026

What this means for you

The same safe pattern works over any data your team needs answers from.

  • Chat with your company data. Plain-English questions over your own databases and reports.
  • Internal AI assistants for operations. Answers for the team without waiting for an analyst.
  • Analytics questions answered without a dashboard. Ask the question instead of building another report.

For a client we would connect it read-only to your PostgreSQL, data warehouse or BI layer, with access rules. See our AI agent development services.

Frequently asked questions

Can the AI change our data?

No. Its access is read-only by design.

Which data can it use?

Whatever it's connected to read-only, such as databases, a warehouse or reports.

How do you control cost?

By choosing the right model per feature.